1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 25d4f98a2SYan Zheng /* 35d4f98a2SYan Zheng * Copyright (C) 2009 Oracle. All rights reserved. 45d4f98a2SYan Zheng */ 55d4f98a2SYan Zheng 65d4f98a2SYan Zheng #include <linux/sched.h> 75d4f98a2SYan Zheng #include <linux/pagemap.h> 85d4f98a2SYan Zheng #include <linux/writeback.h> 95d4f98a2SYan Zheng #include <linux/blkdev.h> 105d4f98a2SYan Zheng #include <linux/rbtree.h> 115a0e3ad6STejun Heo #include <linux/slab.h> 12726a3421SQu Wenruo #include <linux/error-injection.h> 135d4f98a2SYan Zheng #include "ctree.h" 145d4f98a2SYan Zheng #include "disk-io.h" 155d4f98a2SYan Zheng #include "transaction.h" 165d4f98a2SYan Zheng #include "volumes.h" 175d4f98a2SYan Zheng #include "locking.h" 185d4f98a2SYan Zheng #include "btrfs_inode.h" 195d4f98a2SYan Zheng #include "async-thread.h" 200af3d00bSJosef Bacik #include "free-space-cache.h" 21581bb050SLi Zefan #include "inode-map.h" 2262b99540SQu Wenruo #include "qgroup.h" 23cdccee99SLiu Bo #include "print-tree.h" 2486736342SJosef Bacik #include "delalloc-space.h" 25aac0023cSJosef Bacik #include "block-group.h" 265d4f98a2SYan Zheng 275d4f98a2SYan Zheng /* 280c891389SQu Wenruo * Relocation overview 290c891389SQu Wenruo * 300c891389SQu Wenruo * [What does relocation do] 310c891389SQu Wenruo * 320c891389SQu Wenruo * The objective of relocation is to relocate all extents of the target block 330c891389SQu Wenruo * group to other block groups. 340c891389SQu Wenruo * This is utilized by resize (shrink only), profile converting, compacting 350c891389SQu Wenruo * space, or balance routine to spread chunks over devices. 360c891389SQu Wenruo * 370c891389SQu Wenruo * Before | After 380c891389SQu Wenruo * ------------------------------------------------------------------ 390c891389SQu Wenruo * BG A: 10 data extents | BG A: deleted 400c891389SQu Wenruo * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) 410c891389SQu Wenruo * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) 420c891389SQu Wenruo * 430c891389SQu Wenruo * [How does relocation work] 440c891389SQu Wenruo * 450c891389SQu Wenruo * 1. Mark the target block group read-only 460c891389SQu Wenruo * New extents won't be allocated from the target block group. 470c891389SQu Wenruo * 480c891389SQu Wenruo * 2.1 Record each extent in the target block group 490c891389SQu Wenruo * To build a proper map of extents to be relocated. 500c891389SQu Wenruo * 510c891389SQu Wenruo * 2.2 Build data reloc tree and reloc trees 520c891389SQu Wenruo * Data reloc tree will contain an inode, recording all newly relocated 530c891389SQu Wenruo * data extents. 540c891389SQu Wenruo * There will be only one data reloc tree for one data block group. 550c891389SQu Wenruo * 560c891389SQu Wenruo * Reloc tree will be a special snapshot of its source tree, containing 570c891389SQu Wenruo * relocated tree blocks. 580c891389SQu Wenruo * Each tree referring to a tree block in target block group will get its 590c891389SQu Wenruo * reloc tree built. 600c891389SQu Wenruo * 610c891389SQu Wenruo * 2.3 Swap source tree with its corresponding reloc tree 620c891389SQu Wenruo * Each involved tree only refers to new extents after swap. 630c891389SQu Wenruo * 640c891389SQu Wenruo * 3. Cleanup reloc trees and data reloc tree. 650c891389SQu Wenruo * As old extents in the target block group are still referenced by reloc 660c891389SQu Wenruo * trees, we need to clean them up before really freeing the target block 670c891389SQu Wenruo * group. 680c891389SQu Wenruo * 690c891389SQu Wenruo * The main complexity is in steps 2.2 and 2.3. 700c891389SQu Wenruo * 710c891389SQu Wenruo * The entry point of relocation is relocate_block_group() function. 720c891389SQu Wenruo */ 730c891389SQu Wenruo 740c891389SQu Wenruo /* 755d4f98a2SYan Zheng * backref_node, mapping_node and tree_block start with this 765d4f98a2SYan Zheng */ 775d4f98a2SYan Zheng struct tree_entry { 785d4f98a2SYan Zheng struct rb_node rb_node; 795d4f98a2SYan Zheng u64 bytenr; 805d4f98a2SYan Zheng }; 815d4f98a2SYan Zheng 825d4f98a2SYan Zheng /* 835d4f98a2SYan Zheng * present a tree block in the backref cache 845d4f98a2SYan Zheng */ 855d4f98a2SYan Zheng struct backref_node { 865d4f98a2SYan Zheng struct rb_node rb_node; 875d4f98a2SYan Zheng u64 bytenr; 883fd0a558SYan, Zheng 893fd0a558SYan, Zheng u64 new_bytenr; 903fd0a558SYan, Zheng /* objectid of tree block owner, can be not uptodate */ 915d4f98a2SYan Zheng u64 owner; 923fd0a558SYan, Zheng /* link to pending, changed or detached list */ 933fd0a558SYan, Zheng struct list_head list; 945d4f98a2SYan Zheng /* list of upper level blocks reference this block */ 955d4f98a2SYan Zheng struct list_head upper; 965d4f98a2SYan Zheng /* list of child blocks in the cache */ 975d4f98a2SYan Zheng struct list_head lower; 985d4f98a2SYan Zheng /* NULL if this node is not tree root */ 995d4f98a2SYan Zheng struct btrfs_root *root; 1005d4f98a2SYan Zheng /* extent buffer got by COW the block */ 1015d4f98a2SYan Zheng struct extent_buffer *eb; 1025d4f98a2SYan Zheng /* level of tree block */ 1035d4f98a2SYan Zheng unsigned int level:8; 1043fd0a558SYan, Zheng /* is the block in non-reference counted tree */ 1053fd0a558SYan, Zheng unsigned int cowonly:1; 1063fd0a558SYan, Zheng /* 1 if no child node in the cache */ 1075d4f98a2SYan Zheng unsigned int lowest:1; 1085d4f98a2SYan Zheng /* is the extent buffer locked */ 1095d4f98a2SYan Zheng unsigned int locked:1; 1105d4f98a2SYan Zheng /* has the block been processed */ 1115d4f98a2SYan Zheng unsigned int processed:1; 1125d4f98a2SYan Zheng /* have backrefs of this block been checked */ 1135d4f98a2SYan Zheng unsigned int checked:1; 1143fd0a558SYan, Zheng /* 1153fd0a558SYan, Zheng * 1 if corresponding block has been cowed but some upper 1163fd0a558SYan, Zheng * level block pointers may not point to the new location 1173fd0a558SYan, Zheng */ 1183fd0a558SYan, Zheng unsigned int pending:1; 1193fd0a558SYan, Zheng /* 1203fd0a558SYan, Zheng * 1 if the backref node isn't connected to any other 1213fd0a558SYan, Zheng * backref node. 1223fd0a558SYan, Zheng */ 1233fd0a558SYan, Zheng unsigned int detached:1; 1245d4f98a2SYan Zheng }; 1255d4f98a2SYan Zheng 1265d4f98a2SYan Zheng /* 1275d4f98a2SYan Zheng * present a block pointer in the backref cache 1285d4f98a2SYan Zheng */ 1295d4f98a2SYan Zheng struct backref_edge { 1305d4f98a2SYan Zheng struct list_head list[2]; 1315d4f98a2SYan Zheng struct backref_node *node[2]; 1325d4f98a2SYan Zheng }; 1335d4f98a2SYan Zheng 1345d4f98a2SYan Zheng #define LOWER 0 1355d4f98a2SYan Zheng #define UPPER 1 1360647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 1375d4f98a2SYan Zheng 1385d4f98a2SYan Zheng struct backref_cache { 1395d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 1405d4f98a2SYan Zheng struct rb_root rb_root; 1413fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 1423fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 1433fd0a558SYan, Zheng /* 1443fd0a558SYan, Zheng * list of blocks that have been cowed but some block 1453fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 1463fd0a558SYan, Zheng * new location 1473fd0a558SYan, Zheng */ 1485d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 1493fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1503fd0a558SYan, Zheng struct list_head leaves; 1513fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1523fd0a558SYan, Zheng struct list_head changed; 1533fd0a558SYan, Zheng /* list of detached backref node. */ 1543fd0a558SYan, Zheng struct list_head detached; 1553fd0a558SYan, Zheng 1563fd0a558SYan, Zheng u64 last_trans; 1573fd0a558SYan, Zheng 1583fd0a558SYan, Zheng int nr_nodes; 1593fd0a558SYan, Zheng int nr_edges; 1605d4f98a2SYan Zheng }; 1615d4f98a2SYan Zheng 1625d4f98a2SYan Zheng /* 1635d4f98a2SYan Zheng * map address of tree root to tree 1645d4f98a2SYan Zheng */ 1655d4f98a2SYan Zheng struct mapping_node { 1665d4f98a2SYan Zheng struct rb_node rb_node; 1675d4f98a2SYan Zheng u64 bytenr; 1685d4f98a2SYan Zheng void *data; 1695d4f98a2SYan Zheng }; 1705d4f98a2SYan Zheng 1715d4f98a2SYan Zheng struct mapping_tree { 1725d4f98a2SYan Zheng struct rb_root rb_root; 1735d4f98a2SYan Zheng spinlock_t lock; 1745d4f98a2SYan Zheng }; 1755d4f98a2SYan Zheng 1765d4f98a2SYan Zheng /* 1775d4f98a2SYan Zheng * present a tree block to process 1785d4f98a2SYan Zheng */ 1795d4f98a2SYan Zheng struct tree_block { 1805d4f98a2SYan Zheng struct rb_node rb_node; 1815d4f98a2SYan Zheng u64 bytenr; 1825d4f98a2SYan Zheng struct btrfs_key key; 1835d4f98a2SYan Zheng unsigned int level:8; 1845d4f98a2SYan Zheng unsigned int key_ready:1; 1855d4f98a2SYan Zheng }; 1865d4f98a2SYan Zheng 1870257bb82SYan, Zheng #define MAX_EXTENTS 128 1880257bb82SYan, Zheng 1890257bb82SYan, Zheng struct file_extent_cluster { 1900257bb82SYan, Zheng u64 start; 1910257bb82SYan, Zheng u64 end; 1920257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1930257bb82SYan, Zheng unsigned int nr; 1940257bb82SYan, Zheng }; 1950257bb82SYan, Zheng 1965d4f98a2SYan Zheng struct reloc_control { 1975d4f98a2SYan Zheng /* block group to relocate */ 19832da5386SDavid Sterba struct btrfs_block_group *block_group; 1995d4f98a2SYan Zheng /* extent tree */ 2005d4f98a2SYan Zheng struct btrfs_root *extent_root; 2015d4f98a2SYan Zheng /* inode for moving data */ 2025d4f98a2SYan Zheng struct inode *data_inode; 2033fd0a558SYan, Zheng 2043fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 2053fd0a558SYan, Zheng 2063fd0a558SYan, Zheng struct backref_cache backref_cache; 2073fd0a558SYan, Zheng 2083fd0a558SYan, Zheng struct file_extent_cluster cluster; 2095d4f98a2SYan Zheng /* tree blocks have been processed */ 2105d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 2115d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 2125d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 2135d4f98a2SYan Zheng /* list of reloc trees */ 2145d4f98a2SYan Zheng struct list_head reloc_roots; 215d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 216d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 2173fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 2183fd0a558SYan, Zheng u64 merging_rsv_size; 2193fd0a558SYan, Zheng /* size of relocated tree nodes */ 2203fd0a558SYan, Zheng u64 nodes_relocated; 2210647bf56SWang Shilong /* reserved size for block group relocation*/ 2220647bf56SWang Shilong u64 reserved_bytes; 2233fd0a558SYan, Zheng 2245d4f98a2SYan Zheng u64 search_start; 2255d4f98a2SYan Zheng u64 extents_found; 2263fd0a558SYan, Zheng 2273fd0a558SYan, Zheng unsigned int stage:8; 2283fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 2293fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 2305d4f98a2SYan Zheng unsigned int found_file_extent:1; 2315d4f98a2SYan Zheng }; 2325d4f98a2SYan Zheng 2335d4f98a2SYan Zheng /* stages of data relocation */ 2345d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 2355d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 2365d4f98a2SYan Zheng 2373fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 2383fd0a558SYan, Zheng struct backref_node *node); 2393fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 2403fd0a558SYan, Zheng struct backref_node *node); 2415d4f98a2SYan Zheng 2425d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 2435d4f98a2SYan Zheng { 2446bef4d31SEric Paris tree->rb_root = RB_ROOT; 2455d4f98a2SYan Zheng spin_lock_init(&tree->lock); 2465d4f98a2SYan Zheng } 2475d4f98a2SYan Zheng 2485d4f98a2SYan Zheng static void backref_cache_init(struct backref_cache *cache) 2495d4f98a2SYan Zheng { 2505d4f98a2SYan Zheng int i; 2516bef4d31SEric Paris cache->rb_root = RB_ROOT; 2525d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2535d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2543fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2553fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2563fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 2575d4f98a2SYan Zheng } 2585d4f98a2SYan Zheng 2593fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 2605d4f98a2SYan Zheng { 2613fd0a558SYan, Zheng struct backref_node *node; 2623fd0a558SYan, Zheng int i; 2633fd0a558SYan, Zheng 2643fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2653fd0a558SYan, Zheng node = list_entry(cache->detached.next, 2663fd0a558SYan, Zheng struct backref_node, list); 2673fd0a558SYan, Zheng remove_backref_node(cache, node); 2683fd0a558SYan, Zheng } 2693fd0a558SYan, Zheng 2703fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 2713fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 2723fd0a558SYan, Zheng struct backref_node, lower); 2733fd0a558SYan, Zheng remove_backref_node(cache, node); 2743fd0a558SYan, Zheng } 2753fd0a558SYan, Zheng 2763fd0a558SYan, Zheng cache->last_trans = 0; 2773fd0a558SYan, Zheng 2783fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 279f4907095SLiu Bo ASSERT(list_empty(&cache->pending[i])); 280f4907095SLiu Bo ASSERT(list_empty(&cache->changed)); 281f4907095SLiu Bo ASSERT(list_empty(&cache->detached)); 282f4907095SLiu Bo ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); 283f4907095SLiu Bo ASSERT(!cache->nr_nodes); 284f4907095SLiu Bo ASSERT(!cache->nr_edges); 2853fd0a558SYan, Zheng } 2863fd0a558SYan, Zheng 2873fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 2883fd0a558SYan, Zheng { 2893fd0a558SYan, Zheng struct backref_node *node; 2903fd0a558SYan, Zheng 2913fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 2923fd0a558SYan, Zheng if (node) { 2933fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 2945d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 2955d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 2965d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 2973fd0a558SYan, Zheng cache->nr_nodes++; 2983fd0a558SYan, Zheng } 2993fd0a558SYan, Zheng return node; 3003fd0a558SYan, Zheng } 3013fd0a558SYan, Zheng 3023fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 3033fd0a558SYan, Zheng struct backref_node *node) 3043fd0a558SYan, Zheng { 3053fd0a558SYan, Zheng if (node) { 3063fd0a558SYan, Zheng cache->nr_nodes--; 30700246528SJosef Bacik btrfs_put_root(node->root); 3083fd0a558SYan, Zheng kfree(node); 3093fd0a558SYan, Zheng } 3103fd0a558SYan, Zheng } 3113fd0a558SYan, Zheng 3123fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 3133fd0a558SYan, Zheng { 3143fd0a558SYan, Zheng struct backref_edge *edge; 3153fd0a558SYan, Zheng 3163fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 3173fd0a558SYan, Zheng if (edge) 3183fd0a558SYan, Zheng cache->nr_edges++; 3193fd0a558SYan, Zheng return edge; 3203fd0a558SYan, Zheng } 3213fd0a558SYan, Zheng 3223fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 3233fd0a558SYan, Zheng struct backref_edge *edge) 3243fd0a558SYan, Zheng { 3253fd0a558SYan, Zheng if (edge) { 3263fd0a558SYan, Zheng cache->nr_edges--; 3273fd0a558SYan, Zheng kfree(edge); 3283fd0a558SYan, Zheng } 3295d4f98a2SYan Zheng } 3305d4f98a2SYan Zheng 3315d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 3325d4f98a2SYan Zheng struct rb_node *node) 3335d4f98a2SYan Zheng { 3345d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 3355d4f98a2SYan Zheng struct rb_node *parent = NULL; 3365d4f98a2SYan Zheng struct tree_entry *entry; 3375d4f98a2SYan Zheng 3385d4f98a2SYan Zheng while (*p) { 3395d4f98a2SYan Zheng parent = *p; 3405d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 3415d4f98a2SYan Zheng 3425d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3435d4f98a2SYan Zheng p = &(*p)->rb_left; 3445d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3455d4f98a2SYan Zheng p = &(*p)->rb_right; 3465d4f98a2SYan Zheng else 3475d4f98a2SYan Zheng return parent; 3485d4f98a2SYan Zheng } 3495d4f98a2SYan Zheng 3505d4f98a2SYan Zheng rb_link_node(node, parent, p); 3515d4f98a2SYan Zheng rb_insert_color(node, root); 3525d4f98a2SYan Zheng return NULL; 3535d4f98a2SYan Zheng } 3545d4f98a2SYan Zheng 3555d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 3565d4f98a2SYan Zheng { 3575d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 3585d4f98a2SYan Zheng struct tree_entry *entry; 3595d4f98a2SYan Zheng 3605d4f98a2SYan Zheng while (n) { 3615d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 3625d4f98a2SYan Zheng 3635d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3645d4f98a2SYan Zheng n = n->rb_left; 3655d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3665d4f98a2SYan Zheng n = n->rb_right; 3675d4f98a2SYan Zheng else 3685d4f98a2SYan Zheng return n; 3695d4f98a2SYan Zheng } 3705d4f98a2SYan Zheng return NULL; 3715d4f98a2SYan Zheng } 3725d4f98a2SYan Zheng 37348a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 37443c04fb1SJeff Mahoney { 37543c04fb1SJeff Mahoney 37643c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 37743c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 37843c04fb1SJeff Mahoney rb_node); 37943c04fb1SJeff Mahoney if (bnode->root) 38043c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 3815d163e0eSJeff Mahoney btrfs_panic(fs_info, errno, 3825d163e0eSJeff Mahoney "Inconsistency in backref cache found at offset %llu", 3835d163e0eSJeff Mahoney bytenr); 38443c04fb1SJeff Mahoney } 38543c04fb1SJeff Mahoney 3865d4f98a2SYan Zheng /* 3875d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3885d4f98a2SYan Zheng */ 3895d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3905d4f98a2SYan Zheng struct backref_edge *edges[], 3915d4f98a2SYan Zheng int *index) 3925d4f98a2SYan Zheng { 3935d4f98a2SYan Zheng struct backref_edge *edge; 3945d4f98a2SYan Zheng int idx = *index; 3955d4f98a2SYan Zheng 3965d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3975d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3985d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3995d4f98a2SYan Zheng edges[idx++] = edge; 4005d4f98a2SYan Zheng node = edge->node[UPPER]; 4015d4f98a2SYan Zheng } 4023fd0a558SYan, Zheng BUG_ON(node->detached); 4035d4f98a2SYan Zheng *index = idx; 4045d4f98a2SYan Zheng return node; 4055d4f98a2SYan Zheng } 4065d4f98a2SYan Zheng 4075d4f98a2SYan Zheng /* 4085d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 4095d4f98a2SYan Zheng */ 4105d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 4115d4f98a2SYan Zheng int *index) 4125d4f98a2SYan Zheng { 4135d4f98a2SYan Zheng struct backref_edge *edge; 4145d4f98a2SYan Zheng struct backref_node *lower; 4155d4f98a2SYan Zheng int idx = *index; 4165d4f98a2SYan Zheng 4175d4f98a2SYan Zheng while (idx > 0) { 4185d4f98a2SYan Zheng edge = edges[idx - 1]; 4195d4f98a2SYan Zheng lower = edge->node[LOWER]; 4205d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 4215d4f98a2SYan Zheng idx--; 4225d4f98a2SYan Zheng continue; 4235d4f98a2SYan Zheng } 4245d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 4255d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 4265d4f98a2SYan Zheng edges[idx - 1] = edge; 4275d4f98a2SYan Zheng *index = idx; 4285d4f98a2SYan Zheng return edge->node[UPPER]; 4295d4f98a2SYan Zheng } 4305d4f98a2SYan Zheng *index = 0; 4315d4f98a2SYan Zheng return NULL; 4325d4f98a2SYan Zheng } 4335d4f98a2SYan Zheng 4343fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 4355d4f98a2SYan Zheng { 4365d4f98a2SYan Zheng if (node->locked) { 4375d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 4385d4f98a2SYan Zheng node->locked = 0; 4395d4f98a2SYan Zheng } 4403fd0a558SYan, Zheng } 4413fd0a558SYan, Zheng 4423fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 4433fd0a558SYan, Zheng { 4443fd0a558SYan, Zheng if (node->eb) { 4453fd0a558SYan, Zheng unlock_node_buffer(node); 4465d4f98a2SYan Zheng free_extent_buffer(node->eb); 4475d4f98a2SYan Zheng node->eb = NULL; 4485d4f98a2SYan Zheng } 4495d4f98a2SYan Zheng } 4505d4f98a2SYan Zheng 4515d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4525d4f98a2SYan Zheng struct backref_node *node) 4535d4f98a2SYan Zheng { 4545d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4555d4f98a2SYan Zheng 4565d4f98a2SYan Zheng drop_node_buffer(node); 4573fd0a558SYan, Zheng list_del(&node->list); 4585d4f98a2SYan Zheng list_del(&node->lower); 4593fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4605d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4613fd0a558SYan, Zheng free_backref_node(tree, node); 4625d4f98a2SYan Zheng } 4635d4f98a2SYan Zheng 4645d4f98a2SYan Zheng /* 4655d4f98a2SYan Zheng * remove a backref node from the backref cache 4665d4f98a2SYan Zheng */ 4675d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4685d4f98a2SYan Zheng struct backref_node *node) 4695d4f98a2SYan Zheng { 4705d4f98a2SYan Zheng struct backref_node *upper; 4715d4f98a2SYan Zheng struct backref_edge *edge; 4725d4f98a2SYan Zheng 4735d4f98a2SYan Zheng if (!node) 4745d4f98a2SYan Zheng return; 4755d4f98a2SYan Zheng 4763fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4775d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4785d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4795d4f98a2SYan Zheng list[LOWER]); 4805d4f98a2SYan Zheng upper = edge->node[UPPER]; 4815d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4825d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4833fd0a558SYan, Zheng free_backref_edge(cache, edge); 4843fd0a558SYan, Zheng 4853fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4863fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4873fd0a558SYan, Zheng drop_backref_node(cache, node); 4883fd0a558SYan, Zheng node = upper; 4893fd0a558SYan, Zheng node->lowest = 1; 4903fd0a558SYan, Zheng continue; 4913fd0a558SYan, Zheng } 4925d4f98a2SYan Zheng /* 4933fd0a558SYan, Zheng * add the node to leaf node list if no other 4945d4f98a2SYan Zheng * child block cached. 4955d4f98a2SYan Zheng */ 4965d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4973fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4985d4f98a2SYan Zheng upper->lowest = 1; 4995d4f98a2SYan Zheng } 5005d4f98a2SYan Zheng } 5013fd0a558SYan, Zheng 5025d4f98a2SYan Zheng drop_backref_node(cache, node); 5035d4f98a2SYan Zheng } 5045d4f98a2SYan Zheng 5053fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 5063fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 5073fd0a558SYan, Zheng { 5083fd0a558SYan, Zheng struct rb_node *rb_node; 5093fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 5103fd0a558SYan, Zheng node->bytenr = bytenr; 5113fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 51243c04fb1SJeff Mahoney if (rb_node) 51343c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 5143fd0a558SYan, Zheng } 5153fd0a558SYan, Zheng 5163fd0a558SYan, Zheng /* 5173fd0a558SYan, Zheng * update backref cache after a transaction commit 5183fd0a558SYan, Zheng */ 5193fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 5203fd0a558SYan, Zheng struct backref_cache *cache) 5213fd0a558SYan, Zheng { 5223fd0a558SYan, Zheng struct backref_node *node; 5233fd0a558SYan, Zheng int level = 0; 5243fd0a558SYan, Zheng 5253fd0a558SYan, Zheng if (cache->last_trans == 0) { 5263fd0a558SYan, Zheng cache->last_trans = trans->transid; 5273fd0a558SYan, Zheng return 0; 5283fd0a558SYan, Zheng } 5293fd0a558SYan, Zheng 5303fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 5313fd0a558SYan, Zheng return 0; 5323fd0a558SYan, Zheng 5333fd0a558SYan, Zheng /* 5343fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 5353fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 5363fd0a558SYan, Zheng * so the detached nodes are no longer useful. 5373fd0a558SYan, Zheng */ 5383fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 5393fd0a558SYan, Zheng node = list_entry(cache->detached.next, 5403fd0a558SYan, Zheng struct backref_node, list); 5413fd0a558SYan, Zheng remove_backref_node(cache, node); 5423fd0a558SYan, Zheng } 5433fd0a558SYan, Zheng 5443fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 5453fd0a558SYan, Zheng node = list_entry(cache->changed.next, 5463fd0a558SYan, Zheng struct backref_node, list); 5473fd0a558SYan, Zheng list_del_init(&node->list); 5483fd0a558SYan, Zheng BUG_ON(node->pending); 5493fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5503fd0a558SYan, Zheng } 5513fd0a558SYan, Zheng 5523fd0a558SYan, Zheng /* 5533fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5543fd0a558SYan, Zheng * errors during processing the pending nodes. 5553fd0a558SYan, Zheng */ 5563fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5573fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5583fd0a558SYan, Zheng BUG_ON(!node->pending); 5593fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5603fd0a558SYan, Zheng continue; 5613fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5623fd0a558SYan, Zheng } 5633fd0a558SYan, Zheng } 5643fd0a558SYan, Zheng 5653fd0a558SYan, Zheng cache->last_trans = 0; 5663fd0a558SYan, Zheng return 1; 5673fd0a558SYan, Zheng } 5683fd0a558SYan, Zheng 5696282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 5706282675eSQu Wenruo { 5716282675eSQu Wenruo /* 5726282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 5736282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 5746282675eSQu Wenruo * trying to access reloc_root 5756282675eSQu Wenruo */ 5766282675eSQu Wenruo smp_rmb(); 5776282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 5786282675eSQu Wenruo return true; 5796282675eSQu Wenruo return false; 5806282675eSQu Wenruo } 5816282675eSQu Wenruo 5826282675eSQu Wenruo /* 5836282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 5846282675eSQu Wenruo * 5856282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 5866282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 5876282675eSQu Wenruo * from no reloc root. But should_ignore_root() below is a special case. 5886282675eSQu Wenruo */ 5896282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 5906282675eSQu Wenruo { 5916282675eSQu Wenruo if (reloc_root_is_dead(root)) 5926282675eSQu Wenruo return false; 5936282675eSQu Wenruo if (!root->reloc_root) 5946282675eSQu Wenruo return false; 5956282675eSQu Wenruo return true; 5966282675eSQu Wenruo } 597f2a97a9dSDavid Sterba 5983fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5993fd0a558SYan, Zheng { 6003fd0a558SYan, Zheng struct btrfs_root *reloc_root; 6013fd0a558SYan, Zheng 60227cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 6033fd0a558SYan, Zheng return 0; 6043fd0a558SYan, Zheng 6056282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 6066282675eSQu Wenruo if (reloc_root_is_dead(root)) 6076282675eSQu Wenruo return 1; 6086282675eSQu Wenruo 6093fd0a558SYan, Zheng reloc_root = root->reloc_root; 6103fd0a558SYan, Zheng if (!reloc_root) 6113fd0a558SYan, Zheng return 0; 6123fd0a558SYan, Zheng 6133fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 6143fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 6153fd0a558SYan, Zheng return 0; 6163fd0a558SYan, Zheng /* 6173fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 6183fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 6193fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 6203fd0a558SYan, Zheng * relocation. 6213fd0a558SYan, Zheng */ 6223fd0a558SYan, Zheng return 1; 6233fd0a558SYan, Zheng } 6245d4f98a2SYan Zheng /* 6255d4f98a2SYan Zheng * find reloc tree by address of tree root 6265d4f98a2SYan Zheng */ 6275d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 6285d4f98a2SYan Zheng u64 bytenr) 6295d4f98a2SYan Zheng { 6305d4f98a2SYan Zheng struct rb_node *rb_node; 6315d4f98a2SYan Zheng struct mapping_node *node; 6325d4f98a2SYan Zheng struct btrfs_root *root = NULL; 6335d4f98a2SYan Zheng 6345d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 6355d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 6365d4f98a2SYan Zheng if (rb_node) { 6375d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 6385d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 6395d4f98a2SYan Zheng } 6405d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 64100246528SJosef Bacik return btrfs_grab_root(root); 6425d4f98a2SYan Zheng } 6435d4f98a2SYan Zheng 6445d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 6455d4f98a2SYan Zheng u64 root_objectid) 6465d4f98a2SYan Zheng { 6475d4f98a2SYan Zheng struct btrfs_key key; 6485d4f98a2SYan Zheng 6495d4f98a2SYan Zheng key.objectid = root_objectid; 6505d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 6515d4f98a2SYan Zheng key.offset = (u64)-1; 6525d4f98a2SYan Zheng 653bc44d7c4SJosef Bacik return btrfs_get_fs_root(fs_info, &key, false); 6545d4f98a2SYan Zheng } 6555d4f98a2SYan Zheng 6565d4f98a2SYan Zheng static noinline_for_stack 6575d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 6585d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 6595d4f98a2SYan Zheng { 6603173a18fSJosef Bacik struct btrfs_key key; 6615d4f98a2SYan Zheng struct btrfs_extent_item *ei; 6625d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 6635d4f98a2SYan Zheng u32 item_size; 6645d4f98a2SYan Zheng 6653173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6663173a18fSJosef Bacik 6675d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 668ba3c2b19SNikolay Borisov if (item_size < sizeof(*ei)) { 669ba3c2b19SNikolay Borisov btrfs_print_v0_err(leaf->fs_info); 670ba3c2b19SNikolay Borisov btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL); 671ba3c2b19SNikolay Borisov return 1; 672ba3c2b19SNikolay Borisov } 6735d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6745d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6755d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6765d4f98a2SYan Zheng 6773173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6783173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6795d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6805d4f98a2SYan Zheng return 1; 6815d4f98a2SYan Zheng } 682d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 683d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 684d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 685d062d13cSJosef Bacik return 1; 686d062d13cSJosef Bacik } 6875d4f98a2SYan Zheng 6883173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6895d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6905d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6913173a18fSJosef Bacik } else { 6923173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6933173a18fSJosef Bacik } 6945d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6955d4f98a2SYan Zheng return 0; 6965d4f98a2SYan Zheng } 6975d4f98a2SYan Zheng 6985d4f98a2SYan Zheng /* 6995d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 7005d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 7015d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 7025d4f98a2SYan Zheng * 7035d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 70401327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 70501327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 7065d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 7075d4f98a2SYan Zheng * 7085d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 7095d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 7105d4f98a2SYan Zheng * block are also cached. 7115d4f98a2SYan Zheng */ 7123fd0a558SYan, Zheng static noinline_for_stack 7133fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 7145d4f98a2SYan Zheng struct btrfs_key *node_key, 7155d4f98a2SYan Zheng int level, u64 bytenr) 7165d4f98a2SYan Zheng { 7173fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 718fa6ac715SQu Wenruo struct btrfs_path *path1; /* For searching extent root */ 719fa6ac715SQu Wenruo struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */ 7205d4f98a2SYan Zheng struct extent_buffer *eb; 7215d4f98a2SYan Zheng struct btrfs_root *root; 7225d4f98a2SYan Zheng struct backref_node *cur; 7235d4f98a2SYan Zheng struct backref_node *upper; 7245d4f98a2SYan Zheng struct backref_node *lower; 7255d4f98a2SYan Zheng struct backref_node *node = NULL; 7265d4f98a2SYan Zheng struct backref_node *exist = NULL; 7275d4f98a2SYan Zheng struct backref_edge *edge; 7285d4f98a2SYan Zheng struct rb_node *rb_node; 7295d4f98a2SYan Zheng struct btrfs_key key; 7305d4f98a2SYan Zheng unsigned long end; 7315d4f98a2SYan Zheng unsigned long ptr; 732fa6ac715SQu Wenruo LIST_HEAD(list); /* Pending edge list, upper node needs to be checked */ 7333fd0a558SYan, Zheng LIST_HEAD(useless); 7343fd0a558SYan, Zheng int cowonly; 7355d4f98a2SYan Zheng int ret; 7365d4f98a2SYan Zheng int err = 0; 737b6c60c80SJosef Bacik bool need_check = true; 7385d4f98a2SYan Zheng 7395d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 7405d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 7415d4f98a2SYan Zheng if (!path1 || !path2) { 7425d4f98a2SYan Zheng err = -ENOMEM; 7435d4f98a2SYan Zheng goto out; 7445d4f98a2SYan Zheng } 745e4058b54SDavid Sterba path1->reada = READA_FORWARD; 746e4058b54SDavid Sterba path2->reada = READA_FORWARD; 7475d4f98a2SYan Zheng 7483fd0a558SYan, Zheng node = alloc_backref_node(cache); 7495d4f98a2SYan Zheng if (!node) { 7505d4f98a2SYan Zheng err = -ENOMEM; 7515d4f98a2SYan Zheng goto out; 7525d4f98a2SYan Zheng } 7535d4f98a2SYan Zheng 7545d4f98a2SYan Zheng node->bytenr = bytenr; 7555d4f98a2SYan Zheng node->level = level; 7565d4f98a2SYan Zheng node->lowest = 1; 7575d4f98a2SYan Zheng cur = node; 7585d4f98a2SYan Zheng again: 7595d4f98a2SYan Zheng end = 0; 7605d4f98a2SYan Zheng ptr = 0; 7615d4f98a2SYan Zheng key.objectid = cur->bytenr; 7623173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 7635d4f98a2SYan Zheng key.offset = (u64)-1; 7645d4f98a2SYan Zheng 7655d4f98a2SYan Zheng path1->search_commit_root = 1; 7665d4f98a2SYan Zheng path1->skip_locking = 1; 7675d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7685d4f98a2SYan Zheng 0, 0); 7695d4f98a2SYan Zheng if (ret < 0) { 7705d4f98a2SYan Zheng err = ret; 7715d4f98a2SYan Zheng goto out; 7725d4f98a2SYan Zheng } 77375bfb9afSJosef Bacik ASSERT(ret); 77475bfb9afSJosef Bacik ASSERT(path1->slots[0]); 7755d4f98a2SYan Zheng 7765d4f98a2SYan Zheng path1->slots[0]--; 7775d4f98a2SYan Zheng 7785d4f98a2SYan Zheng WARN_ON(cur->checked); 7795d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7805d4f98a2SYan Zheng /* 78170f23fd6SJustin P. Mattock * the backref was added previously when processing 7825d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7835d4f98a2SYan Zheng */ 78475bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7855d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7865d4f98a2SYan Zheng list[LOWER]); 78775bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7885d4f98a2SYan Zheng exist = edge->node[UPPER]; 7895d4f98a2SYan Zheng /* 7905d4f98a2SYan Zheng * add the upper level block to pending list if we need 7915d4f98a2SYan Zheng * check its backrefs 7925d4f98a2SYan Zheng */ 7935d4f98a2SYan Zheng if (!exist->checked) 7945d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7955d4f98a2SYan Zheng } else { 7965d4f98a2SYan Zheng exist = NULL; 7975d4f98a2SYan Zheng } 7985d4f98a2SYan Zheng 7995d4f98a2SYan Zheng while (1) { 8005d4f98a2SYan Zheng cond_resched(); 8015d4f98a2SYan Zheng eb = path1->nodes[0]; 8025d4f98a2SYan Zheng 8035d4f98a2SYan Zheng if (ptr >= end) { 8045d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 8055d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 8065d4f98a2SYan Zheng if (ret < 0) { 8075d4f98a2SYan Zheng err = ret; 8085d4f98a2SYan Zheng goto out; 8095d4f98a2SYan Zheng } 8105d4f98a2SYan Zheng if (ret > 0) 8115d4f98a2SYan Zheng break; 8125d4f98a2SYan Zheng eb = path1->nodes[0]; 8135d4f98a2SYan Zheng } 8145d4f98a2SYan Zheng 8155d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 8165d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 8175d4f98a2SYan Zheng WARN_ON(exist); 8185d4f98a2SYan Zheng break; 8195d4f98a2SYan Zheng } 8205d4f98a2SYan Zheng 8213173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 8223173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 8235d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 8245d4f98a2SYan Zheng &ptr, &end); 8255d4f98a2SYan Zheng if (ret) 8265d4f98a2SYan Zheng goto next; 8275d4f98a2SYan Zheng } 8285d4f98a2SYan Zheng } 8295d4f98a2SYan Zheng 8305d4f98a2SYan Zheng if (ptr < end) { 8315d4f98a2SYan Zheng /* update key for inline back ref */ 8325d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 8333de28d57SLiu Bo int type; 8345d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 8353de28d57SLiu Bo type = btrfs_get_extent_inline_ref_type(eb, iref, 8363de28d57SLiu Bo BTRFS_REF_TYPE_BLOCK); 8373de28d57SLiu Bo if (type == BTRFS_REF_TYPE_INVALID) { 838af431dcbSSu Yue err = -EUCLEAN; 8393de28d57SLiu Bo goto out; 8403de28d57SLiu Bo } 8413de28d57SLiu Bo key.type = type; 8425d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 8433de28d57SLiu Bo 8445d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 8455d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 8465d4f98a2SYan Zheng } 8475d4f98a2SYan Zheng 848fa6ac715SQu Wenruo /* 849fa6ac715SQu Wenruo * Parent node found and matches current inline ref, no need to 850fa6ac715SQu Wenruo * rebuild this node for this inline ref. 851fa6ac715SQu Wenruo */ 8525d4f98a2SYan Zheng if (exist && 8535d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 8545d4f98a2SYan Zheng exist->owner == key.offset) || 8555d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 8565d4f98a2SYan Zheng exist->bytenr == key.offset))) { 8575d4f98a2SYan Zheng exist = NULL; 8585d4f98a2SYan Zheng goto next; 8595d4f98a2SYan Zheng } 8605d4f98a2SYan Zheng 861fa6ac715SQu Wenruo /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ 8625d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8635d4f98a2SYan Zheng if (key.objectid == key.offset) { 8645d4f98a2SYan Zheng /* 865fa6ac715SQu Wenruo * Only root blocks of reloc trees use backref 866fa6ac715SQu Wenruo * pointing to itself. 8675d4f98a2SYan Zheng */ 8685d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 86975bfb9afSJosef Bacik ASSERT(root); 8705d4f98a2SYan Zheng cur->root = root; 8715d4f98a2SYan Zheng break; 8725d4f98a2SYan Zheng } 8735d4f98a2SYan Zheng 8743fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8755d4f98a2SYan Zheng if (!edge) { 8765d4f98a2SYan Zheng err = -ENOMEM; 8775d4f98a2SYan Zheng goto out; 8785d4f98a2SYan Zheng } 8795d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8805d4f98a2SYan Zheng if (!rb_node) { 8813fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8825d4f98a2SYan Zheng if (!upper) { 8833fd0a558SYan, Zheng free_backref_edge(cache, edge); 8845d4f98a2SYan Zheng err = -ENOMEM; 8855d4f98a2SYan Zheng goto out; 8865d4f98a2SYan Zheng } 8875d4f98a2SYan Zheng upper->bytenr = key.offset; 8885d4f98a2SYan Zheng upper->level = cur->level + 1; 8895d4f98a2SYan Zheng /* 8905d4f98a2SYan Zheng * backrefs for the upper level block isn't 8915d4f98a2SYan Zheng * cached, add the block to pending list 8925d4f98a2SYan Zheng */ 8935d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8945d4f98a2SYan Zheng } else { 8955d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8965d4f98a2SYan Zheng rb_node); 89775bfb9afSJosef Bacik ASSERT(upper->checked); 8985d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8995d4f98a2SYan Zheng } 9003fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 9015d4f98a2SYan Zheng edge->node[LOWER] = cur; 9023fd0a558SYan, Zheng edge->node[UPPER] = upper; 9035d4f98a2SYan Zheng 9045d4f98a2SYan Zheng goto next; 9056d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 906ba3c2b19SNikolay Borisov err = -EINVAL; 907ba3c2b19SNikolay Borisov btrfs_print_v0_err(rc->extent_root->fs_info); 908ba3c2b19SNikolay Borisov btrfs_handle_fs_error(rc->extent_root->fs_info, err, 909ba3c2b19SNikolay Borisov NULL); 910ba3c2b19SNikolay Borisov goto out; 9115d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 9125d4f98a2SYan Zheng goto next; 9135d4f98a2SYan Zheng } 9145d4f98a2SYan Zheng 915fa6ac715SQu Wenruo /* 916fa6ac715SQu Wenruo * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset 917fa6ac715SQu Wenruo * means the root objectid. We need to search the tree to get 918fa6ac715SQu Wenruo * its parent bytenr. 919fa6ac715SQu Wenruo */ 9205d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 9215d4f98a2SYan Zheng if (IS_ERR(root)) { 9225d4f98a2SYan Zheng err = PTR_ERR(root); 9235d4f98a2SYan Zheng goto out; 9245d4f98a2SYan Zheng } 9255d4f98a2SYan Zheng 92627cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 9273fd0a558SYan, Zheng cur->cowonly = 1; 9283fd0a558SYan, Zheng 9295d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 9305d4f98a2SYan Zheng /* tree root */ 93175bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9325d4f98a2SYan Zheng cur->bytenr); 9330b530bc5SJosef Bacik if (should_ignore_root(root)) { 93400246528SJosef Bacik btrfs_put_root(root); 9353fd0a558SYan, Zheng list_add(&cur->list, &useless); 9360b530bc5SJosef Bacik } else { 9375d4f98a2SYan Zheng cur->root = root; 9380b530bc5SJosef Bacik } 9395d4f98a2SYan Zheng break; 9405d4f98a2SYan Zheng } 9415d4f98a2SYan Zheng 9425d4f98a2SYan Zheng level = cur->level + 1; 9435d4f98a2SYan Zheng 944fa6ac715SQu Wenruo /* Search the tree to find parent blocks referring the block. */ 9455d4f98a2SYan Zheng path2->search_commit_root = 1; 9465d4f98a2SYan Zheng path2->skip_locking = 1; 9475d4f98a2SYan Zheng path2->lowest_level = level; 9485d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 9495d4f98a2SYan Zheng path2->lowest_level = 0; 9505d4f98a2SYan Zheng if (ret < 0) { 95100246528SJosef Bacik btrfs_put_root(root); 9525d4f98a2SYan Zheng err = ret; 9535d4f98a2SYan Zheng goto out; 9545d4f98a2SYan Zheng } 95533c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 95633c66f43SYan Zheng path2->slots[level]--; 9575d4f98a2SYan Zheng 9585d4f98a2SYan Zheng eb = path2->nodes[level]; 9593561b9dbSLiu Bo if (btrfs_node_blockptr(eb, path2->slots[level]) != 9603561b9dbSLiu Bo cur->bytenr) { 9613561b9dbSLiu Bo btrfs_err(root->fs_info, 9623561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 9634fd786e6SMisono Tomohiro cur->bytenr, level - 1, 9644fd786e6SMisono Tomohiro root->root_key.objectid, 9653561b9dbSLiu Bo node_key->objectid, node_key->type, 9663561b9dbSLiu Bo node_key->offset); 96700246528SJosef Bacik btrfs_put_root(root); 9683561b9dbSLiu Bo err = -ENOENT; 9693561b9dbSLiu Bo goto out; 9703561b9dbSLiu Bo } 9715d4f98a2SYan Zheng lower = cur; 972b6c60c80SJosef Bacik need_check = true; 973fa6ac715SQu Wenruo 974fa6ac715SQu Wenruo /* Add all nodes and edges in the path */ 9755d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9765d4f98a2SYan Zheng if (!path2->nodes[level]) { 97775bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9785d4f98a2SYan Zheng lower->bytenr); 9790b530bc5SJosef Bacik if (should_ignore_root(root)) { 98000246528SJosef Bacik btrfs_put_root(root); 9813fd0a558SYan, Zheng list_add(&lower->list, &useless); 9820b530bc5SJosef Bacik } else { 9835d4f98a2SYan Zheng lower->root = root; 9840b530bc5SJosef Bacik } 9855d4f98a2SYan Zheng break; 9865d4f98a2SYan Zheng } 9875d4f98a2SYan Zheng 9883fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9895d4f98a2SYan Zheng if (!edge) { 99000246528SJosef Bacik btrfs_put_root(root); 9915d4f98a2SYan Zheng err = -ENOMEM; 9925d4f98a2SYan Zheng goto out; 9935d4f98a2SYan Zheng } 9945d4f98a2SYan Zheng 9955d4f98a2SYan Zheng eb = path2->nodes[level]; 9965d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9975d4f98a2SYan Zheng if (!rb_node) { 9983fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9995d4f98a2SYan Zheng if (!upper) { 100000246528SJosef Bacik btrfs_put_root(root); 10013fd0a558SYan, Zheng free_backref_edge(cache, edge); 10025d4f98a2SYan Zheng err = -ENOMEM; 10035d4f98a2SYan Zheng goto out; 10045d4f98a2SYan Zheng } 10055d4f98a2SYan Zheng upper->bytenr = eb->start; 10065d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 10075d4f98a2SYan Zheng upper->level = lower->level + 1; 100827cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 100927cdeb70SMiao Xie &root->state)) 10103fd0a558SYan, Zheng upper->cowonly = 1; 10115d4f98a2SYan Zheng 10125d4f98a2SYan Zheng /* 10135d4f98a2SYan Zheng * if we know the block isn't shared 10145d4f98a2SYan Zheng * we can void checking its backrefs. 10155d4f98a2SYan Zheng */ 10165d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 10175d4f98a2SYan Zheng upper->checked = 0; 10185d4f98a2SYan Zheng else 10195d4f98a2SYan Zheng upper->checked = 1; 10205d4f98a2SYan Zheng 10215d4f98a2SYan Zheng /* 10225d4f98a2SYan Zheng * add the block to pending list if we 1023b6c60c80SJosef Bacik * need check its backrefs, we only do this once 1024b6c60c80SJosef Bacik * while walking up a tree as we will catch 1025b6c60c80SJosef Bacik * anything else later on. 10265d4f98a2SYan Zheng */ 1027b6c60c80SJosef Bacik if (!upper->checked && need_check) { 1028b6c60c80SJosef Bacik need_check = false; 10295d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 10305d4f98a2SYan Zheng &list); 1031bbe90514SJosef Bacik } else { 1032bbe90514SJosef Bacik if (upper->checked) 1033bbe90514SJosef Bacik need_check = true; 10345d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 1035bbe90514SJosef Bacik } 10365d4f98a2SYan Zheng } else { 10375d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 10385d4f98a2SYan Zheng rb_node); 103975bfb9afSJosef Bacik ASSERT(upper->checked); 10405d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 10413fd0a558SYan, Zheng if (!upper->owner) 10423fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 10435d4f98a2SYan Zheng } 10445d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 10455d4f98a2SYan Zheng edge->node[LOWER] = lower; 10463fd0a558SYan, Zheng edge->node[UPPER] = upper; 10475d4f98a2SYan Zheng 10480b530bc5SJosef Bacik if (rb_node) { 104900246528SJosef Bacik btrfs_put_root(root); 10505d4f98a2SYan Zheng break; 10510b530bc5SJosef Bacik } 10525d4f98a2SYan Zheng lower = upper; 10535d4f98a2SYan Zheng upper = NULL; 10545d4f98a2SYan Zheng } 1055b3b4aa74SDavid Sterba btrfs_release_path(path2); 10565d4f98a2SYan Zheng next: 10575d4f98a2SYan Zheng if (ptr < end) { 10585d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 10595d4f98a2SYan Zheng if (ptr >= end) { 10605d4f98a2SYan Zheng WARN_ON(ptr > end); 10615d4f98a2SYan Zheng ptr = 0; 10625d4f98a2SYan Zheng end = 0; 10635d4f98a2SYan Zheng } 10645d4f98a2SYan Zheng } 10655d4f98a2SYan Zheng if (ptr >= end) 10665d4f98a2SYan Zheng path1->slots[0]++; 10675d4f98a2SYan Zheng } 1068b3b4aa74SDavid Sterba btrfs_release_path(path1); 10695d4f98a2SYan Zheng 10705d4f98a2SYan Zheng cur->checked = 1; 10715d4f98a2SYan Zheng WARN_ON(exist); 10725d4f98a2SYan Zheng 10735d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10745d4f98a2SYan Zheng if (!list_empty(&list)) { 10755d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10765d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10775d4f98a2SYan Zheng cur = edge->node[UPPER]; 10785d4f98a2SYan Zheng goto again; 10795d4f98a2SYan Zheng } 10805d4f98a2SYan Zheng 10815d4f98a2SYan Zheng /* 10825d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10835d4f98a2SYan Zheng * into the cache. 10845d4f98a2SYan Zheng */ 108575bfb9afSJosef Bacik ASSERT(node->checked); 10863fd0a558SYan, Zheng cowonly = node->cowonly; 10873fd0a558SYan, Zheng if (!cowonly) { 10883fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10893fd0a558SYan, Zheng &node->rb_node); 109043c04fb1SJeff Mahoney if (rb_node) 109143c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10923fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10933fd0a558SYan, Zheng } 10945d4f98a2SYan Zheng 10955d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10965d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10975d4f98a2SYan Zheng 10985d4f98a2SYan Zheng while (!list_empty(&list)) { 10995d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 11005d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 11015d4f98a2SYan Zheng upper = edge->node[UPPER]; 11023fd0a558SYan, Zheng if (upper->detached) { 11033fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11043fd0a558SYan, Zheng lower = edge->node[LOWER]; 11053fd0a558SYan, Zheng free_backref_edge(cache, edge); 11063fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11073fd0a558SYan, Zheng list_add(&lower->list, &useless); 11083fd0a558SYan, Zheng continue; 11093fd0a558SYan, Zheng } 11105d4f98a2SYan Zheng 11115d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 11125d4f98a2SYan Zheng if (upper->lowest) { 11135d4f98a2SYan Zheng list_del_init(&upper->lower); 11145d4f98a2SYan Zheng upper->lowest = 0; 11155d4f98a2SYan Zheng } 11165d4f98a2SYan Zheng 11175d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 11185d4f98a2SYan Zheng continue; 11195d4f98a2SYan Zheng } 11205d4f98a2SYan Zheng 112175bfb9afSJosef Bacik if (!upper->checked) { 112275bfb9afSJosef Bacik /* 112375bfb9afSJosef Bacik * Still want to blow up for developers since this is a 112475bfb9afSJosef Bacik * logic bug. 112575bfb9afSJosef Bacik */ 112675bfb9afSJosef Bacik ASSERT(0); 112775bfb9afSJosef Bacik err = -EINVAL; 112875bfb9afSJosef Bacik goto out; 112975bfb9afSJosef Bacik } 113075bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 113175bfb9afSJosef Bacik ASSERT(0); 113275bfb9afSJosef Bacik err = -EINVAL; 113375bfb9afSJosef Bacik goto out; 113475bfb9afSJosef Bacik } 113575bfb9afSJosef Bacik 11363fd0a558SYan, Zheng if (!cowonly) { 11375d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 11385d4f98a2SYan Zheng &upper->rb_node); 113943c04fb1SJeff Mahoney if (rb_node) 114043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 114143c04fb1SJeff Mahoney upper->bytenr); 11423fd0a558SYan, Zheng } 11435d4f98a2SYan Zheng 11445d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 11455d4f98a2SYan Zheng 11465d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 11475d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 11485d4f98a2SYan Zheng } 11493fd0a558SYan, Zheng /* 11503fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 11513fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 11523fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 11533fd0a558SYan, Zheng * lookup. 11543fd0a558SYan, Zheng */ 11553fd0a558SYan, Zheng while (!list_empty(&useless)) { 11563fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 11573fd0a558SYan, Zheng list_del_init(&upper->list); 115875bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 11593fd0a558SYan, Zheng if (upper == node) 11603fd0a558SYan, Zheng node = NULL; 11613fd0a558SYan, Zheng if (upper->lowest) { 11623fd0a558SYan, Zheng list_del_init(&upper->lower); 11633fd0a558SYan, Zheng upper->lowest = 0; 11643fd0a558SYan, Zheng } 11653fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 11663fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 11673fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 11683fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 11693fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11703fd0a558SYan, Zheng lower = edge->node[LOWER]; 11713fd0a558SYan, Zheng free_backref_edge(cache, edge); 11723fd0a558SYan, Zheng 11733fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11743fd0a558SYan, Zheng list_add(&lower->list, &useless); 11753fd0a558SYan, Zheng } 11763fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11773fd0a558SYan, Zheng if (upper->level > 0) { 11783fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11793fd0a558SYan, Zheng upper->detached = 1; 11803fd0a558SYan, Zheng } else { 11813fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11823fd0a558SYan, Zheng free_backref_node(cache, upper); 11833fd0a558SYan, Zheng } 11843fd0a558SYan, Zheng } 11855d4f98a2SYan Zheng out: 11865d4f98a2SYan Zheng btrfs_free_path(path1); 11875d4f98a2SYan Zheng btrfs_free_path(path2); 11885d4f98a2SYan Zheng if (err) { 11893fd0a558SYan, Zheng while (!list_empty(&useless)) { 11903fd0a558SYan, Zheng lower = list_entry(useless.next, 119175bfb9afSJosef Bacik struct backref_node, list); 119275bfb9afSJosef Bacik list_del_init(&lower->list); 11933fd0a558SYan, Zheng } 119475bfb9afSJosef Bacik while (!list_empty(&list)) { 119575bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 119675bfb9afSJosef Bacik list[UPPER]); 119775bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11983fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 119975bfb9afSJosef Bacik lower = edge->node[LOWER]; 12005d4f98a2SYan Zheng upper = edge->node[UPPER]; 12013fd0a558SYan, Zheng free_backref_edge(cache, edge); 120275bfb9afSJosef Bacik 120375bfb9afSJosef Bacik /* 120475bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 120575bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 120675bfb9afSJosef Bacik */ 120775bfb9afSJosef Bacik if (list_empty(&lower->upper) && 120875bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 120975bfb9afSJosef Bacik list_add(&lower->list, &useless); 121075bfb9afSJosef Bacik 121175bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 121275bfb9afSJosef Bacik continue; 121375bfb9afSJosef Bacik 121401327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 121575bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 121675bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 121775bfb9afSJosef Bacik if (list_empty(&upper->upper)) 121875bfb9afSJosef Bacik list_add(&upper->list, &useless); 121975bfb9afSJosef Bacik } 122075bfb9afSJosef Bacik 122175bfb9afSJosef Bacik while (!list_empty(&useless)) { 122275bfb9afSJosef Bacik lower = list_entry(useless.next, 122375bfb9afSJosef Bacik struct backref_node, list); 122475bfb9afSJosef Bacik list_del_init(&lower->list); 12250fd8c3daSLiu Bo if (lower == node) 12260fd8c3daSLiu Bo node = NULL; 122775bfb9afSJosef Bacik free_backref_node(cache, lower); 12285d4f98a2SYan Zheng } 12290fd8c3daSLiu Bo 12300fd8c3daSLiu Bo free_backref_node(cache, node); 12315d4f98a2SYan Zheng return ERR_PTR(err); 12325d4f98a2SYan Zheng } 123375bfb9afSJosef Bacik ASSERT(!node || !node->detached); 12345d4f98a2SYan Zheng return node; 12355d4f98a2SYan Zheng } 12365d4f98a2SYan Zheng 12375d4f98a2SYan Zheng /* 12383fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 12393fd0a558SYan, Zheng * the backref node is created by cloning backref node that 12403fd0a558SYan, Zheng * corresponds to root of source tree 12413fd0a558SYan, Zheng */ 12423fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 12433fd0a558SYan, Zheng struct reloc_control *rc, 12443fd0a558SYan, Zheng struct btrfs_root *src, 12453fd0a558SYan, Zheng struct btrfs_root *dest) 12463fd0a558SYan, Zheng { 12473fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 12483fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 12493fd0a558SYan, Zheng struct backref_node *node = NULL; 12503fd0a558SYan, Zheng struct backref_node *new_node; 12513fd0a558SYan, Zheng struct backref_edge *edge; 12523fd0a558SYan, Zheng struct backref_edge *new_edge; 12533fd0a558SYan, Zheng struct rb_node *rb_node; 12543fd0a558SYan, Zheng 12553fd0a558SYan, Zheng if (cache->last_trans > 0) 12563fd0a558SYan, Zheng update_backref_cache(trans, cache); 12573fd0a558SYan, Zheng 12583fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 12593fd0a558SYan, Zheng if (rb_node) { 12603fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 12613fd0a558SYan, Zheng if (node->detached) 12623fd0a558SYan, Zheng node = NULL; 12633fd0a558SYan, Zheng else 12643fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 12653fd0a558SYan, Zheng } 12663fd0a558SYan, Zheng 12673fd0a558SYan, Zheng if (!node) { 12683fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 12693fd0a558SYan, Zheng reloc_root->commit_root->start); 12703fd0a558SYan, Zheng if (rb_node) { 12713fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 12723fd0a558SYan, Zheng rb_node); 12733fd0a558SYan, Zheng BUG_ON(node->detached); 12743fd0a558SYan, Zheng } 12753fd0a558SYan, Zheng } 12763fd0a558SYan, Zheng 12773fd0a558SYan, Zheng if (!node) 12783fd0a558SYan, Zheng return 0; 12793fd0a558SYan, Zheng 12803fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12813fd0a558SYan, Zheng if (!new_node) 12823fd0a558SYan, Zheng return -ENOMEM; 12833fd0a558SYan, Zheng 12843fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12853fd0a558SYan, Zheng new_node->level = node->level; 12863fd0a558SYan, Zheng new_node->lowest = node->lowest; 12876848ad64SYan, Zheng new_node->checked = 1; 128800246528SJosef Bacik new_node->root = btrfs_grab_root(dest); 12890b530bc5SJosef Bacik ASSERT(new_node->root); 12903fd0a558SYan, Zheng 12913fd0a558SYan, Zheng if (!node->lowest) { 12923fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12933fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12943fd0a558SYan, Zheng if (!new_edge) 12953fd0a558SYan, Zheng goto fail; 12963fd0a558SYan, Zheng 12973fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12983fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12993fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 13003fd0a558SYan, Zheng &new_node->lower); 13013fd0a558SYan, Zheng } 130276b9e23dSMiao Xie } else { 130376b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 13043fd0a558SYan, Zheng } 13053fd0a558SYan, Zheng 13063fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 13073fd0a558SYan, Zheng &new_node->rb_node); 130843c04fb1SJeff Mahoney if (rb_node) 130943c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 13103fd0a558SYan, Zheng 13113fd0a558SYan, Zheng if (!new_node->lowest) { 13123fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 13133fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 13143fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 13153fd0a558SYan, Zheng } 13163fd0a558SYan, Zheng } 13173fd0a558SYan, Zheng return 0; 13183fd0a558SYan, Zheng fail: 13193fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 13203fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 13213fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 13223fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 13233fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 13243fd0a558SYan, Zheng } 13253fd0a558SYan, Zheng free_backref_node(cache, new_node); 13263fd0a558SYan, Zheng return -ENOMEM; 13273fd0a558SYan, Zheng } 13283fd0a558SYan, Zheng 13293fd0a558SYan, Zheng /* 13305d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 13315d4f98a2SYan Zheng */ 1332ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 13335d4f98a2SYan Zheng { 13340b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13355d4f98a2SYan Zheng struct rb_node *rb_node; 13365d4f98a2SYan Zheng struct mapping_node *node; 13370b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 13385d4f98a2SYan Zheng 13395d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1340ffd7b339SJeff Mahoney if (!node) 1341ffd7b339SJeff Mahoney return -ENOMEM; 13425d4f98a2SYan Zheng 13435d4f98a2SYan Zheng node->bytenr = root->node->start; 13445d4f98a2SYan Zheng node->data = root; 13455d4f98a2SYan Zheng 13465d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13475d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13485d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13495d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1350ffd7b339SJeff Mahoney if (rb_node) { 13510b246afaSJeff Mahoney btrfs_panic(fs_info, -EEXIST, 13525d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 13535d163e0eSJeff Mahoney node->bytenr); 1354ffd7b339SJeff Mahoney } 13555d4f98a2SYan Zheng 13565d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 13575d4f98a2SYan Zheng return 0; 13585d4f98a2SYan Zheng } 13595d4f98a2SYan Zheng 13605d4f98a2SYan Zheng /* 1361c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 13625d4f98a2SYan Zheng * mapping 13635d4f98a2SYan Zheng */ 1364c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 13655d4f98a2SYan Zheng { 13660b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13675d4f98a2SYan Zheng struct rb_node *rb_node; 13685d4f98a2SYan Zheng struct mapping_node *node = NULL; 13690b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 13705d4f98a2SYan Zheng 137165c6e82bSQu Wenruo if (rc && root->node) { 13725d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13735d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1374c974c464SWang Shilong root->node->start); 1375c974c464SWang Shilong if (rb_node) { 1376c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1377c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1378c974c464SWang Shilong } 1379c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1380c974c464SWang Shilong if (!node) 1381c974c464SWang Shilong return; 1382c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1383389305b2SQu Wenruo } 1384c974c464SWang Shilong 13850b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1386c974c464SWang Shilong list_del_init(&root->root_list); 13870b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1388c974c464SWang Shilong kfree(node); 1389c974c464SWang Shilong } 1390c974c464SWang Shilong 1391c974c464SWang Shilong /* 1392c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1393c974c464SWang Shilong * mapping 1394c974c464SWang Shilong */ 1395c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1396c974c464SWang Shilong { 13970b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1398c974c464SWang Shilong struct rb_node *rb_node; 1399c974c464SWang Shilong struct mapping_node *node = NULL; 14000b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1401c974c464SWang Shilong 1402c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1403c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1404c974c464SWang Shilong root->node->start); 14055d4f98a2SYan Zheng if (rb_node) { 14065d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 14075d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 14085d4f98a2SYan Zheng } 14095d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 14105d4f98a2SYan Zheng 14118f71f3e0SLiu Bo if (!node) 14128f71f3e0SLiu Bo return 0; 14135d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 14145d4f98a2SYan Zheng 14155d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1416c974c464SWang Shilong node->bytenr = new_bytenr; 14175d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 14185d4f98a2SYan Zheng node->bytenr, &node->rb_node); 14195d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 142043c04fb1SJeff Mahoney if (rb_node) 142143c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 14225d4f98a2SYan Zheng return 0; 14235d4f98a2SYan Zheng } 14245d4f98a2SYan Zheng 14253fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 14263fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 14275d4f98a2SYan Zheng { 14280b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14295d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14305d4f98a2SYan Zheng struct extent_buffer *eb; 14315d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14325d4f98a2SYan Zheng struct btrfs_key root_key; 14335d4f98a2SYan Zheng int ret; 14345d4f98a2SYan Zheng 14355d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 14365d4f98a2SYan Zheng BUG_ON(!root_item); 14375d4f98a2SYan Zheng 14385d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 14395d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 14403fd0a558SYan, Zheng root_key.offset = objectid; 14415d4f98a2SYan Zheng 14423fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 1443054570a1SFilipe Manana u64 commit_root_gen; 1444054570a1SFilipe Manana 14453fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 14465d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 14475d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 14485d4f98a2SYan Zheng BUG_ON(ret); 1449054570a1SFilipe Manana /* 1450054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 1451054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 1452054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 1453054570a1SFilipe Manana * either inside the critical section of a transaction commit 1454054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 1455054570a1SFilipe Manana * it's created before the transaction commit is started. 1456054570a1SFilipe Manana */ 1457054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 1458054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 14593fd0a558SYan, Zheng } else { 14603fd0a558SYan, Zheng /* 14613fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 14623fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 14633fd0a558SYan, Zheng * modified after it was created have RELOC flag 14643fd0a558SYan, Zheng * set in their headers. so it's OK to not update 14653fd0a558SYan, Zheng * the 'last_snapshot'. 14663fd0a558SYan, Zheng */ 14673fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 14683fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 14693fd0a558SYan, Zheng BUG_ON(ret); 14703fd0a558SYan, Zheng } 14713fd0a558SYan, Zheng 14725d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 14735d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 14745d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14755d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14763fd0a558SYan, Zheng 14773fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14783fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14793fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14803fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14815d4f98a2SYan Zheng root_item->drop_level = 0; 14823fd0a558SYan, Zheng } 14835d4f98a2SYan Zheng 14845d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14855d4f98a2SYan Zheng free_extent_buffer(eb); 14865d4f98a2SYan Zheng 14870b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 14885d4f98a2SYan Zheng &root_key, root_item); 14895d4f98a2SYan Zheng BUG_ON(ret); 14905d4f98a2SYan Zheng kfree(root_item); 14915d4f98a2SYan Zheng 14923dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); 14935d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14943dbf1738SJosef Bacik set_bit(BTRFS_ROOT_REF_COWS, &reloc_root->state); 14955d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14963fd0a558SYan, Zheng return reloc_root; 14973fd0a558SYan, Zheng } 14983fd0a558SYan, Zheng 14993fd0a558SYan, Zheng /* 15003fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 15013fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 15023fd0a558SYan, Zheng */ 15033fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 15043fd0a558SYan, Zheng struct btrfs_root *root) 15053fd0a558SYan, Zheng { 15060b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15073fd0a558SYan, Zheng struct btrfs_root *reloc_root; 15080b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 150920dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 15103fd0a558SYan, Zheng int clear_rsv = 0; 1511ffd7b339SJeff Mahoney int ret; 15123fd0a558SYan, Zheng 15131fac4a54SQu Wenruo /* 15141fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 15151fac4a54SQu Wenruo * create/update the dead reloc tree 15161fac4a54SQu Wenruo */ 15176282675eSQu Wenruo if (reloc_root_is_dead(root)) 15181fac4a54SQu Wenruo return 0; 15191fac4a54SQu Wenruo 15203fd0a558SYan, Zheng if (root->reloc_root) { 15213fd0a558SYan, Zheng reloc_root = root->reloc_root; 15223fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 15233fd0a558SYan, Zheng return 0; 15243fd0a558SYan, Zheng } 15253fd0a558SYan, Zheng 15263fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 15273fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 15283fd0a558SYan, Zheng return 0; 15293fd0a558SYan, Zheng 153020dd2cbfSMiao Xie if (!trans->reloc_reserved) { 153120dd2cbfSMiao Xie rsv = trans->block_rsv; 15323fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 15333fd0a558SYan, Zheng clear_rsv = 1; 15343fd0a558SYan, Zheng } 15353fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 15363fd0a558SYan, Zheng if (clear_rsv) 153720dd2cbfSMiao Xie trans->block_rsv = rsv; 15385d4f98a2SYan Zheng 1539ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1540ffd7b339SJeff Mahoney BUG_ON(ret < 0); 15415d4f98a2SYan Zheng root->reloc_root = reloc_root; 15425d4f98a2SYan Zheng return 0; 15435d4f98a2SYan Zheng } 15445d4f98a2SYan Zheng 15455d4f98a2SYan Zheng /* 15465d4f98a2SYan Zheng * update root item of reloc tree 15475d4f98a2SYan Zheng */ 15485d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 15495d4f98a2SYan Zheng struct btrfs_root *root) 15505d4f98a2SYan Zheng { 15510b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15525d4f98a2SYan Zheng struct btrfs_root *reloc_root; 15535d4f98a2SYan Zheng struct btrfs_root_item *root_item; 15545d4f98a2SYan Zheng int ret; 15555d4f98a2SYan Zheng 15566282675eSQu Wenruo if (!have_reloc_root(root)) 15577585717fSChris Mason goto out; 15585d4f98a2SYan Zheng 15595d4f98a2SYan Zheng reloc_root = root->reloc_root; 15605d4f98a2SYan Zheng root_item = &reloc_root->root_item; 15615d4f98a2SYan Zheng 1562d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 15630b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 15643fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 1565d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 15666282675eSQu Wenruo /* 15676282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 15686282675eSQu Wenruo * have_reloc_root will not touch it from now on. 15696282675eSQu Wenruo */ 15706282675eSQu Wenruo smp_wmb(); 1571c974c464SWang Shilong __del_reloc_root(reloc_root); 15725d4f98a2SYan Zheng } 15735d4f98a2SYan Zheng 15745d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 15755d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 15765d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 15775d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 15785d4f98a2SYan Zheng } 15795d4f98a2SYan Zheng 15800b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 15815d4f98a2SYan Zheng &reloc_root->root_key, root_item); 15825d4f98a2SYan Zheng BUG_ON(ret); 15837585717fSChris Mason 15847585717fSChris Mason out: 15855d4f98a2SYan Zheng return 0; 15865d4f98a2SYan Zheng } 15875d4f98a2SYan Zheng 15885d4f98a2SYan Zheng /* 15895d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15905d4f98a2SYan Zheng * in a subvolume 15915d4f98a2SYan Zheng */ 15925d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15935d4f98a2SYan Zheng { 15945d4f98a2SYan Zheng struct rb_node *node; 15955d4f98a2SYan Zheng struct rb_node *prev; 15965d4f98a2SYan Zheng struct btrfs_inode *entry; 15975d4f98a2SYan Zheng struct inode *inode; 15985d4f98a2SYan Zheng 15995d4f98a2SYan Zheng spin_lock(&root->inode_lock); 16005d4f98a2SYan Zheng again: 16015d4f98a2SYan Zheng node = root->inode_tree.rb_node; 16025d4f98a2SYan Zheng prev = NULL; 16035d4f98a2SYan Zheng while (node) { 16045d4f98a2SYan Zheng prev = node; 16055d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 16065d4f98a2SYan Zheng 16074a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 16085d4f98a2SYan Zheng node = node->rb_left; 16094a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 16105d4f98a2SYan Zheng node = node->rb_right; 16115d4f98a2SYan Zheng else 16125d4f98a2SYan Zheng break; 16135d4f98a2SYan Zheng } 16145d4f98a2SYan Zheng if (!node) { 16155d4f98a2SYan Zheng while (prev) { 16165d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 16174a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 16185d4f98a2SYan Zheng node = prev; 16195d4f98a2SYan Zheng break; 16205d4f98a2SYan Zheng } 16215d4f98a2SYan Zheng prev = rb_next(prev); 16225d4f98a2SYan Zheng } 16235d4f98a2SYan Zheng } 16245d4f98a2SYan Zheng while (node) { 16255d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 16265d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 16275d4f98a2SYan Zheng if (inode) { 16285d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 16295d4f98a2SYan Zheng return inode; 16305d4f98a2SYan Zheng } 16315d4f98a2SYan Zheng 16324a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 16335d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 16345d4f98a2SYan Zheng goto again; 16355d4f98a2SYan Zheng 16365d4f98a2SYan Zheng node = rb_next(node); 16375d4f98a2SYan Zheng } 16385d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 16395d4f98a2SYan Zheng return NULL; 16405d4f98a2SYan Zheng } 16415d4f98a2SYan Zheng 164232da5386SDavid Sterba static int in_block_group(u64 bytenr, struct btrfs_block_group *block_group) 16435d4f98a2SYan Zheng { 1644b3470b5dSDavid Sterba if (bytenr >= block_group->start && 1645b3470b5dSDavid Sterba bytenr < block_group->start + block_group->length) 16465d4f98a2SYan Zheng return 1; 16475d4f98a2SYan Zheng return 0; 16485d4f98a2SYan Zheng } 16495d4f98a2SYan Zheng 16505d4f98a2SYan Zheng /* 16515d4f98a2SYan Zheng * get new location of data 16525d4f98a2SYan Zheng */ 16535d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 16545d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 16555d4f98a2SYan Zheng { 16565d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 16575d4f98a2SYan Zheng struct btrfs_path *path; 16585d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16595d4f98a2SYan Zheng struct extent_buffer *leaf; 16605d4f98a2SYan Zheng int ret; 16615d4f98a2SYan Zheng 16625d4f98a2SYan Zheng path = btrfs_alloc_path(); 16635d4f98a2SYan Zheng if (!path) 16645d4f98a2SYan Zheng return -ENOMEM; 16655d4f98a2SYan Zheng 16665d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1667f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1668f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 16695d4f98a2SYan Zheng if (ret < 0) 16705d4f98a2SYan Zheng goto out; 16715d4f98a2SYan Zheng if (ret > 0) { 16725d4f98a2SYan Zheng ret = -ENOENT; 16735d4f98a2SYan Zheng goto out; 16745d4f98a2SYan Zheng } 16755d4f98a2SYan Zheng 16765d4f98a2SYan Zheng leaf = path->nodes[0]; 16775d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 16785d4f98a2SYan Zheng struct btrfs_file_extent_item); 16795d4f98a2SYan Zheng 16805d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 16815d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 16825d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 16835d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 16845d4f98a2SYan Zheng 16855d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 168683d4cfd4SJosef Bacik ret = -EINVAL; 16875d4f98a2SYan Zheng goto out; 16885d4f98a2SYan Zheng } 16895d4f98a2SYan Zheng 16905d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16915d4f98a2SYan Zheng ret = 0; 16925d4f98a2SYan Zheng out: 16935d4f98a2SYan Zheng btrfs_free_path(path); 16945d4f98a2SYan Zheng return ret; 16955d4f98a2SYan Zheng } 16965d4f98a2SYan Zheng 16975d4f98a2SYan Zheng /* 16985d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 16995d4f98a2SYan Zheng * the new locations. 17005d4f98a2SYan Zheng */ 17013fd0a558SYan, Zheng static noinline_for_stack 17023fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 17035d4f98a2SYan Zheng struct reloc_control *rc, 17045d4f98a2SYan Zheng struct btrfs_root *root, 17053fd0a558SYan, Zheng struct extent_buffer *leaf) 17065d4f98a2SYan Zheng { 17070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 17085d4f98a2SYan Zheng struct btrfs_key key; 17095d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 17105d4f98a2SYan Zheng struct inode *inode = NULL; 17115d4f98a2SYan Zheng u64 parent; 17125d4f98a2SYan Zheng u64 bytenr; 17133fd0a558SYan, Zheng u64 new_bytenr = 0; 17145d4f98a2SYan Zheng u64 num_bytes; 17155d4f98a2SYan Zheng u64 end; 17165d4f98a2SYan Zheng u32 nritems; 17175d4f98a2SYan Zheng u32 i; 171883d4cfd4SJosef Bacik int ret = 0; 17195d4f98a2SYan Zheng int first = 1; 17205d4f98a2SYan Zheng int dirty = 0; 17215d4f98a2SYan Zheng 17225d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 17235d4f98a2SYan Zheng return 0; 17245d4f98a2SYan Zheng 17255d4f98a2SYan Zheng /* reloc trees always use full backref */ 17265d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 17275d4f98a2SYan Zheng parent = leaf->start; 17285d4f98a2SYan Zheng else 17295d4f98a2SYan Zheng parent = 0; 17305d4f98a2SYan Zheng 17315d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 17325d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 173382fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 173482fa113fSQu Wenruo 17355d4f98a2SYan Zheng cond_resched(); 17365d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 17375d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 17385d4f98a2SYan Zheng continue; 17395d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 17405d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 17415d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 17425d4f98a2SYan Zheng continue; 17435d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 17445d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 17455d4f98a2SYan Zheng if (bytenr == 0) 17465d4f98a2SYan Zheng continue; 17475d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 17485d4f98a2SYan Zheng continue; 17495d4f98a2SYan Zheng 17505d4f98a2SYan Zheng /* 17515d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 17525d4f98a2SYan Zheng * to complete and drop the extent cache 17535d4f98a2SYan Zheng */ 17545d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 17555d4f98a2SYan Zheng if (first) { 17565d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 17575d4f98a2SYan Zheng first = 0; 17584a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 17593fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 17605d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 17615d4f98a2SYan Zheng } 17624a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 17635d4f98a2SYan Zheng end = key.offset + 17645d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 17655d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 17660b246afaSJeff Mahoney fs_info->sectorsize)); 17670b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 17685d4f98a2SYan Zheng end--; 17695d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1770d0082371SJeff Mahoney key.offset, end); 17715d4f98a2SYan Zheng if (!ret) 17725d4f98a2SYan Zheng continue; 17735d4f98a2SYan Zheng 1774dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1775dcdbc059SNikolay Borisov key.offset, end, 1); 17765d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1777d0082371SJeff Mahoney key.offset, end); 17785d4f98a2SYan Zheng } 17795d4f98a2SYan Zheng } 17805d4f98a2SYan Zheng 17815d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 17825d4f98a2SYan Zheng bytenr, num_bytes); 178383d4cfd4SJosef Bacik if (ret) { 178483d4cfd4SJosef Bacik /* 178583d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 178683d4cfd4SJosef Bacik * in the file extent yet. 178783d4cfd4SJosef Bacik */ 178883d4cfd4SJosef Bacik break; 17893fd0a558SYan, Zheng } 17905d4f98a2SYan Zheng 17915d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17925d4f98a2SYan Zheng dirty = 1; 17935d4f98a2SYan Zheng 17945d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 179582fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 179682fa113fSQu Wenruo num_bytes, parent); 179782fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 179882fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1799b06c4bf5SFilipe Manana key.objectid, key.offset); 180082fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 180183d4cfd4SJosef Bacik if (ret) { 180266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 180383d4cfd4SJosef Bacik break; 180483d4cfd4SJosef Bacik } 18055d4f98a2SYan Zheng 1806ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1807ffd4bb2aSQu Wenruo num_bytes, parent); 1808ffd4bb2aSQu Wenruo ref.real_root = root->root_key.objectid; 1809ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1810b06c4bf5SFilipe Manana key.objectid, key.offset); 1811ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 181283d4cfd4SJosef Bacik if (ret) { 181366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 181483d4cfd4SJosef Bacik break; 181583d4cfd4SJosef Bacik } 18165d4f98a2SYan Zheng } 18175d4f98a2SYan Zheng if (dirty) 18185d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 18193fd0a558SYan, Zheng if (inode) 18203fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 182183d4cfd4SJosef Bacik return ret; 18225d4f98a2SYan Zheng } 18235d4f98a2SYan Zheng 18245d4f98a2SYan Zheng static noinline_for_stack 18255d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 18265d4f98a2SYan Zheng struct btrfs_path *path, int level) 18275d4f98a2SYan Zheng { 18285d4f98a2SYan Zheng struct btrfs_disk_key key1; 18295d4f98a2SYan Zheng struct btrfs_disk_key key2; 18305d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 18315d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 18325d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 18335d4f98a2SYan Zheng } 18345d4f98a2SYan Zheng 18355d4f98a2SYan Zheng /* 18365d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 18375d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 18385d4f98a2SYan Zheng * reloc tree was create can be replaced. 18395d4f98a2SYan Zheng * 18405d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 18415d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 18425d4f98a2SYan Zheng * errors, a negative error number is returned. 18435d4f98a2SYan Zheng */ 18443fd0a558SYan, Zheng static noinline_for_stack 18453d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 18465d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 18475d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 18485d4f98a2SYan Zheng int lowest_level, int max_level) 18495d4f98a2SYan Zheng { 18500b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 18515d4f98a2SYan Zheng struct extent_buffer *eb; 18525d4f98a2SYan Zheng struct extent_buffer *parent; 185382fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 18545d4f98a2SYan Zheng struct btrfs_key key; 18555d4f98a2SYan Zheng u64 old_bytenr; 18565d4f98a2SYan Zheng u64 new_bytenr; 18575d4f98a2SYan Zheng u64 old_ptr_gen; 18585d4f98a2SYan Zheng u64 new_ptr_gen; 18595d4f98a2SYan Zheng u64 last_snapshot; 18605d4f98a2SYan Zheng u32 blocksize; 18613fd0a558SYan, Zheng int cow = 0; 18625d4f98a2SYan Zheng int level; 18635d4f98a2SYan Zheng int ret; 18645d4f98a2SYan Zheng int slot; 18655d4f98a2SYan Zheng 18665d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 18675d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 18685d4f98a2SYan Zheng 18695d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 18703fd0a558SYan, Zheng again: 18715d4f98a2SYan Zheng slot = path->slots[lowest_level]; 18725d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 18735d4f98a2SYan Zheng 18745d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 18758bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18765d4f98a2SYan Zheng level = btrfs_header_level(eb); 18775d4f98a2SYan Zheng 18785d4f98a2SYan Zheng if (level < lowest_level) { 18795d4f98a2SYan Zheng btrfs_tree_unlock(eb); 18805d4f98a2SYan Zheng free_extent_buffer(eb); 18815d4f98a2SYan Zheng return 0; 18825d4f98a2SYan Zheng } 18835d4f98a2SYan Zheng 18843fd0a558SYan, Zheng if (cow) { 18855d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 18865d4f98a2SYan Zheng BUG_ON(ret); 18873fd0a558SYan, Zheng } 18888bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18895d4f98a2SYan Zheng 18905d4f98a2SYan Zheng if (next_key) { 18915d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18925d4f98a2SYan Zheng next_key->type = (u8)-1; 18935d4f98a2SYan Zheng next_key->offset = (u64)-1; 18945d4f98a2SYan Zheng } 18955d4f98a2SYan Zheng 18965d4f98a2SYan Zheng parent = eb; 18975d4f98a2SYan Zheng while (1) { 1898581c1760SQu Wenruo struct btrfs_key first_key; 1899581c1760SQu Wenruo 19005d4f98a2SYan Zheng level = btrfs_header_level(parent); 19015d4f98a2SYan Zheng BUG_ON(level < lowest_level); 19025d4f98a2SYan Zheng 19035d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 1904cbca7d59SFilipe Manana if (ret < 0) 1905cbca7d59SFilipe Manana break; 19065d4f98a2SYan Zheng if (ret && slot > 0) 19075d4f98a2SYan Zheng slot--; 19085d4f98a2SYan Zheng 19095d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 19105d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 19115d4f98a2SYan Zheng 19125d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 19130b246afaSJeff Mahoney blocksize = fs_info->nodesize; 19145d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 191517515f1bSQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 19165d4f98a2SYan Zheng 19175d4f98a2SYan Zheng if (level <= max_level) { 19185d4f98a2SYan Zheng eb = path->nodes[level]; 19195d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 19205d4f98a2SYan Zheng path->slots[level]); 19215d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 19225d4f98a2SYan Zheng path->slots[level]); 19235d4f98a2SYan Zheng } else { 19245d4f98a2SYan Zheng new_bytenr = 0; 19255d4f98a2SYan Zheng new_ptr_gen = 0; 19265d4f98a2SYan Zheng } 19275d4f98a2SYan Zheng 1928fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 19295d4f98a2SYan Zheng ret = level; 19305d4f98a2SYan Zheng break; 19315d4f98a2SYan Zheng } 19325d4f98a2SYan Zheng 19335d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 19345d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 19353fd0a558SYan, Zheng if (level <= lowest_level) { 19365d4f98a2SYan Zheng ret = 0; 19375d4f98a2SYan Zheng break; 19385d4f98a2SYan Zheng } 19395d4f98a2SYan Zheng 1940581c1760SQu Wenruo eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen, 1941581c1760SQu Wenruo level - 1, &first_key); 194264c043deSLiu Bo if (IS_ERR(eb)) { 194364c043deSLiu Bo ret = PTR_ERR(eb); 1944264813acSLiu Bo break; 194564c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 194664c043deSLiu Bo ret = -EIO; 1947416bc658SJosef Bacik free_extent_buffer(eb); 1948379cde74SStefan Behrens break; 1949416bc658SJosef Bacik } 19505d4f98a2SYan Zheng btrfs_tree_lock(eb); 19513fd0a558SYan, Zheng if (cow) { 19525d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 19535d4f98a2SYan Zheng slot, &eb); 19545d4f98a2SYan Zheng BUG_ON(ret); 19555d4f98a2SYan Zheng } 19568bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 19575d4f98a2SYan Zheng 19585d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19595d4f98a2SYan Zheng free_extent_buffer(parent); 19605d4f98a2SYan Zheng 19615d4f98a2SYan Zheng parent = eb; 19625d4f98a2SYan Zheng continue; 19635d4f98a2SYan Zheng } 19645d4f98a2SYan Zheng 19653fd0a558SYan, Zheng if (!cow) { 19663fd0a558SYan, Zheng btrfs_tree_unlock(parent); 19673fd0a558SYan, Zheng free_extent_buffer(parent); 19683fd0a558SYan, Zheng cow = 1; 19693fd0a558SYan, Zheng goto again; 19703fd0a558SYan, Zheng } 19713fd0a558SYan, Zheng 19725d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 19735d4f98a2SYan Zheng path->slots[level]); 1974b3b4aa74SDavid Sterba btrfs_release_path(path); 19755d4f98a2SYan Zheng 19765d4f98a2SYan Zheng path->lowest_level = level; 19775d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 19785d4f98a2SYan Zheng path->lowest_level = 0; 19795d4f98a2SYan Zheng BUG_ON(ret); 19805d4f98a2SYan Zheng 19815d4f98a2SYan Zheng /* 1982824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1983824d8dffSQu Wenruo * 1984824d8dffSQu Wenruo * We must trace both trees. 1985824d8dffSQu Wenruo * 1) Tree reloc subtree 1986824d8dffSQu Wenruo * If not traced, we will leak data numbers 1987824d8dffSQu Wenruo * 2) Fs subtree 1988824d8dffSQu Wenruo * If not traced, we will double count old data 1989f616f5cdSQu Wenruo * 1990f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1991f616f5cdSQu Wenruo * the swapped tree blocks. 1992f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1993f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1994824d8dffSQu Wenruo */ 1995370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1996370a11b8SQu Wenruo rc->block_group, parent, slot, 1997370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1998370a11b8SQu Wenruo last_snapshot); 1999370a11b8SQu Wenruo if (ret < 0) 2000370a11b8SQu Wenruo break; 2001824d8dffSQu Wenruo /* 20025d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 20035d4f98a2SYan Zheng */ 20045d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 20055d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 20065d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 20075d4f98a2SYan Zheng 20085d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 20095d4f98a2SYan Zheng path->slots[level], old_bytenr); 20105d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 20115d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 20125d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 20135d4f98a2SYan Zheng 201482fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 201582fa113fSQu Wenruo blocksize, path->nodes[level]->start); 201682fa113fSQu Wenruo ref.skip_qgroup = true; 201782fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 201882fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 20195d4f98a2SYan Zheng BUG_ON(ret); 202082fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 202182fa113fSQu Wenruo blocksize, 0); 202282fa113fSQu Wenruo ref.skip_qgroup = true; 202382fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 202482fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 20255d4f98a2SYan Zheng BUG_ON(ret); 20265d4f98a2SYan Zheng 2027ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 2028ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 2029ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 2030ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 2031ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 20325d4f98a2SYan Zheng BUG_ON(ret); 20335d4f98a2SYan Zheng 2034ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 2035ffd4bb2aSQu Wenruo blocksize, 0); 2036ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 2037ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 2038ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 20395d4f98a2SYan Zheng BUG_ON(ret); 20405d4f98a2SYan Zheng 20415d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 20425d4f98a2SYan Zheng 20435d4f98a2SYan Zheng ret = level; 20445d4f98a2SYan Zheng break; 20455d4f98a2SYan Zheng } 20465d4f98a2SYan Zheng btrfs_tree_unlock(parent); 20475d4f98a2SYan Zheng free_extent_buffer(parent); 20485d4f98a2SYan Zheng return ret; 20495d4f98a2SYan Zheng } 20505d4f98a2SYan Zheng 20515d4f98a2SYan Zheng /* 20525d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 20535d4f98a2SYan Zheng */ 20545d4f98a2SYan Zheng static noinline_for_stack 20555d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20565d4f98a2SYan Zheng int *level) 20575d4f98a2SYan Zheng { 20585d4f98a2SYan Zheng struct extent_buffer *eb; 20595d4f98a2SYan Zheng int i; 20605d4f98a2SYan Zheng u64 last_snapshot; 20615d4f98a2SYan Zheng u32 nritems; 20625d4f98a2SYan Zheng 20635d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20645d4f98a2SYan Zheng 20655d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 20665d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20675d4f98a2SYan Zheng path->nodes[i] = NULL; 20685d4f98a2SYan Zheng } 20695d4f98a2SYan Zheng 20705d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 20715d4f98a2SYan Zheng eb = path->nodes[i]; 20725d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20735d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 20745d4f98a2SYan Zheng path->slots[i]++; 20755d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 20765d4f98a2SYan Zheng last_snapshot) 20775d4f98a2SYan Zheng continue; 20785d4f98a2SYan Zheng 20795d4f98a2SYan Zheng *level = i; 20805d4f98a2SYan Zheng return 0; 20815d4f98a2SYan Zheng } 20825d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20835d4f98a2SYan Zheng path->nodes[i] = NULL; 20845d4f98a2SYan Zheng } 20855d4f98a2SYan Zheng return 1; 20865d4f98a2SYan Zheng } 20875d4f98a2SYan Zheng 20885d4f98a2SYan Zheng /* 20895d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 20905d4f98a2SYan Zheng */ 20915d4f98a2SYan Zheng static noinline_for_stack 20925d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20935d4f98a2SYan Zheng int *level) 20945d4f98a2SYan Zheng { 20952ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20965d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 20975d4f98a2SYan Zheng int i; 20985d4f98a2SYan Zheng u64 bytenr; 20995d4f98a2SYan Zheng u64 ptr_gen = 0; 21005d4f98a2SYan Zheng u64 last_snapshot; 21015d4f98a2SYan Zheng u32 nritems; 21025d4f98a2SYan Zheng 21035d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 21045d4f98a2SYan Zheng 21055d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 2106581c1760SQu Wenruo struct btrfs_key first_key; 2107581c1760SQu Wenruo 21085d4f98a2SYan Zheng eb = path->nodes[i]; 21095d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 21105d4f98a2SYan Zheng while (path->slots[i] < nritems) { 21115d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 21125d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 21135d4f98a2SYan Zheng break; 21145d4f98a2SYan Zheng path->slots[i]++; 21155d4f98a2SYan Zheng } 21165d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 21175d4f98a2SYan Zheng if (i == *level) 21185d4f98a2SYan Zheng break; 21195d4f98a2SYan Zheng *level = i + 1; 21205d4f98a2SYan Zheng return 0; 21215d4f98a2SYan Zheng } 21225d4f98a2SYan Zheng if (i == 1) { 21235d4f98a2SYan Zheng *level = i; 21245d4f98a2SYan Zheng return 0; 21255d4f98a2SYan Zheng } 21265d4f98a2SYan Zheng 21275d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2128581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]); 2129581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1, 2130581c1760SQu Wenruo &first_key); 213164c043deSLiu Bo if (IS_ERR(eb)) { 213264c043deSLiu Bo return PTR_ERR(eb); 213364c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2134416bc658SJosef Bacik free_extent_buffer(eb); 2135416bc658SJosef Bacik return -EIO; 2136416bc658SJosef Bacik } 21375d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 21385d4f98a2SYan Zheng path->nodes[i - 1] = eb; 21395d4f98a2SYan Zheng path->slots[i - 1] = 0; 21405d4f98a2SYan Zheng } 21415d4f98a2SYan Zheng return 1; 21425d4f98a2SYan Zheng } 21435d4f98a2SYan Zheng 21445d4f98a2SYan Zheng /* 21455d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 21465d4f98a2SYan Zheng * [min_key, max_key) 21475d4f98a2SYan Zheng */ 21485d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 21495d4f98a2SYan Zheng struct btrfs_key *min_key, 21505d4f98a2SYan Zheng struct btrfs_key *max_key) 21515d4f98a2SYan Zheng { 21520b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 21535d4f98a2SYan Zheng struct inode *inode = NULL; 21545d4f98a2SYan Zheng u64 objectid; 21555d4f98a2SYan Zheng u64 start, end; 215633345d01SLi Zefan u64 ino; 21575d4f98a2SYan Zheng 21585d4f98a2SYan Zheng objectid = min_key->objectid; 21595d4f98a2SYan Zheng while (1) { 21605d4f98a2SYan Zheng cond_resched(); 21615d4f98a2SYan Zheng iput(inode); 21625d4f98a2SYan Zheng 21635d4f98a2SYan Zheng if (objectid > max_key->objectid) 21645d4f98a2SYan Zheng break; 21655d4f98a2SYan Zheng 21665d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 21675d4f98a2SYan Zheng if (!inode) 21685d4f98a2SYan Zheng break; 21694a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 21705d4f98a2SYan Zheng 217133345d01SLi Zefan if (ino > max_key->objectid) { 21725d4f98a2SYan Zheng iput(inode); 21735d4f98a2SYan Zheng break; 21745d4f98a2SYan Zheng } 21755d4f98a2SYan Zheng 217633345d01SLi Zefan objectid = ino + 1; 21775d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 21785d4f98a2SYan Zheng continue; 21795d4f98a2SYan Zheng 218033345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 21815d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 21825d4f98a2SYan Zheng continue; 21835d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 21845d4f98a2SYan Zheng start = 0; 21855d4f98a2SYan Zheng else { 21865d4f98a2SYan Zheng start = min_key->offset; 21870b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 21885d4f98a2SYan Zheng } 21895d4f98a2SYan Zheng } else { 21905d4f98a2SYan Zheng start = 0; 21915d4f98a2SYan Zheng } 21925d4f98a2SYan Zheng 219333345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 21945d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 21955d4f98a2SYan Zheng continue; 21965d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 21975d4f98a2SYan Zheng end = (u64)-1; 21985d4f98a2SYan Zheng } else { 21995d4f98a2SYan Zheng if (max_key->offset == 0) 22005d4f98a2SYan Zheng continue; 22015d4f98a2SYan Zheng end = max_key->offset; 22020b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 22035d4f98a2SYan Zheng end--; 22045d4f98a2SYan Zheng } 22055d4f98a2SYan Zheng } else { 22065d4f98a2SYan Zheng end = (u64)-1; 22075d4f98a2SYan Zheng } 22085d4f98a2SYan Zheng 22095d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2210d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2211dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2212d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 22135d4f98a2SYan Zheng } 22145d4f98a2SYan Zheng return 0; 22155d4f98a2SYan Zheng } 22165d4f98a2SYan Zheng 22175d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 22185d4f98a2SYan Zheng struct btrfs_key *key) 22195d4f98a2SYan Zheng 22205d4f98a2SYan Zheng { 22215d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 22225d4f98a2SYan Zheng if (!path->nodes[level]) 22235d4f98a2SYan Zheng break; 22245d4f98a2SYan Zheng if (path->slots[level] + 1 < 22255d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 22265d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 22275d4f98a2SYan Zheng path->slots[level] + 1); 22285d4f98a2SYan Zheng return 0; 22295d4f98a2SYan Zheng } 22305d4f98a2SYan Zheng level++; 22315d4f98a2SYan Zheng } 22325d4f98a2SYan Zheng return 1; 22335d4f98a2SYan Zheng } 22345d4f98a2SYan Zheng 22355d4f98a2SYan Zheng /* 2236d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 2237d2311e69SQu Wenruo */ 2238d2311e69SQu Wenruo static void insert_dirty_subvol(struct btrfs_trans_handle *trans, 2239d2311e69SQu Wenruo struct reloc_control *rc, 2240d2311e69SQu Wenruo struct btrfs_root *root) 2241d2311e69SQu Wenruo { 2242d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2243d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 2244d2311e69SQu Wenruo 2245d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 2246d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 2247d2311e69SQu Wenruo ASSERT(reloc_root); 2248d2311e69SQu Wenruo 2249d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 2250d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 2251d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 2252d2311e69SQu Wenruo reloc_root_item->drop_level = 0; 2253d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 2254d2311e69SQu Wenruo btrfs_update_reloc_root(trans, root); 2255d2311e69SQu Wenruo 2256d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 225700246528SJosef Bacik btrfs_grab_root(root); 2258d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 2259d2311e69SQu Wenruo } 2260d2311e69SQu Wenruo } 2261d2311e69SQu Wenruo 2262d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 2263d2311e69SQu Wenruo { 2264d2311e69SQu Wenruo struct btrfs_root *root; 2265d2311e69SQu Wenruo struct btrfs_root *next; 2266d2311e69SQu Wenruo int ret = 0; 226730d40577SQu Wenruo int ret2; 2268d2311e69SQu Wenruo 2269d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 2270d2311e69SQu Wenruo reloc_dirty_list) { 227130d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 227230d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 2273d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2274d2311e69SQu Wenruo 2275d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 2276d2311e69SQu Wenruo root->reloc_root = NULL; 2277d2311e69SQu Wenruo if (reloc_root) { 2278d2311e69SQu Wenruo 2279d2311e69SQu Wenruo ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1); 2280d2311e69SQu Wenruo if (ret2 < 0 && !ret) 2281d2311e69SQu Wenruo ret = ret2; 2282d2311e69SQu Wenruo } 22836282675eSQu Wenruo /* 22846282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 22856282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 22866282675eSQu Wenruo */ 22876282675eSQu Wenruo smp_wmb(); 22881fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 228900246528SJosef Bacik btrfs_put_root(root); 229030d40577SQu Wenruo } else { 229130d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 229230d40577SQu Wenruo ret2 = btrfs_drop_snapshot(root, NULL, 0, 1); 229330d40577SQu Wenruo if (ret2 < 0 && !ret) 229430d40577SQu Wenruo ret = ret2; 229530d40577SQu Wenruo } 2296d2311e69SQu Wenruo } 2297d2311e69SQu Wenruo return ret; 2298d2311e69SQu Wenruo } 2299d2311e69SQu Wenruo 2300d2311e69SQu Wenruo /* 23015d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 23025d4f98a2SYan Zheng * fs tree. 23035d4f98a2SYan Zheng */ 23045d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 23055d4f98a2SYan Zheng struct btrfs_root *root) 23065d4f98a2SYan Zheng { 23070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 23085d4f98a2SYan Zheng struct btrfs_key key; 23095d4f98a2SYan Zheng struct btrfs_key next_key; 23109e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 23115d4f98a2SYan Zheng struct btrfs_root *reloc_root; 23125d4f98a2SYan Zheng struct btrfs_root_item *root_item; 23135d4f98a2SYan Zheng struct btrfs_path *path; 23143fd0a558SYan, Zheng struct extent_buffer *leaf; 23155d4f98a2SYan Zheng int level; 23165d4f98a2SYan Zheng int max_level; 23175d4f98a2SYan Zheng int replaced = 0; 23185d4f98a2SYan Zheng int ret; 23195d4f98a2SYan Zheng int err = 0; 23203fd0a558SYan, Zheng u32 min_reserved; 23215d4f98a2SYan Zheng 23225d4f98a2SYan Zheng path = btrfs_alloc_path(); 23235d4f98a2SYan Zheng if (!path) 23245d4f98a2SYan Zheng return -ENOMEM; 2325e4058b54SDavid Sterba path->reada = READA_FORWARD; 23265d4f98a2SYan Zheng 23275d4f98a2SYan Zheng reloc_root = root->reloc_root; 23285d4f98a2SYan Zheng root_item = &reloc_root->root_item; 23295d4f98a2SYan Zheng 23305d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 23315d4f98a2SYan Zheng level = btrfs_root_level(root_item); 233267439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 23335d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 23345d4f98a2SYan Zheng path->slots[level] = 0; 23355d4f98a2SYan Zheng } else { 23365d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 23375d4f98a2SYan Zheng 23385d4f98a2SYan Zheng level = root_item->drop_level; 23395d4f98a2SYan Zheng BUG_ON(level == 0); 23405d4f98a2SYan Zheng path->lowest_level = level; 23415d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 234233c66f43SYan Zheng path->lowest_level = 0; 23435d4f98a2SYan Zheng if (ret < 0) { 23445d4f98a2SYan Zheng btrfs_free_path(path); 23455d4f98a2SYan Zheng return ret; 23465d4f98a2SYan Zheng } 23475d4f98a2SYan Zheng 23485d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 23495d4f98a2SYan Zheng path->slots[level]); 23505d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 23515d4f98a2SYan Zheng 23525d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 23535d4f98a2SYan Zheng } 23545d4f98a2SYan Zheng 23550b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 23565d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 23575d4f98a2SYan Zheng 23585d4f98a2SYan Zheng while (1) { 235908e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 236008e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 23613fd0a558SYan, Zheng if (ret) { 23629e6a0c52SJosef Bacik err = ret; 23639e6a0c52SJosef Bacik goto out; 23643fd0a558SYan, Zheng } 23659e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 23669e6a0c52SJosef Bacik if (IS_ERR(trans)) { 23679e6a0c52SJosef Bacik err = PTR_ERR(trans); 23689e6a0c52SJosef Bacik trans = NULL; 23699e6a0c52SJosef Bacik goto out; 23709e6a0c52SJosef Bacik } 23719e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 23723fd0a558SYan, Zheng 23733fd0a558SYan, Zheng replaced = 0; 23745d4f98a2SYan Zheng max_level = level; 23755d4f98a2SYan Zheng 23765d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 23775d4f98a2SYan Zheng if (ret < 0) { 23785d4f98a2SYan Zheng err = ret; 23795d4f98a2SYan Zheng goto out; 23805d4f98a2SYan Zheng } 23815d4f98a2SYan Zheng if (ret > 0) 23825d4f98a2SYan Zheng break; 23835d4f98a2SYan Zheng 23845d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 23855d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 23865d4f98a2SYan Zheng ret = 0; 23875d4f98a2SYan Zheng } else { 23883d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 23893fd0a558SYan, Zheng &next_key, level, max_level); 23905d4f98a2SYan Zheng } 23915d4f98a2SYan Zheng if (ret < 0) { 23925d4f98a2SYan Zheng err = ret; 23935d4f98a2SYan Zheng goto out; 23945d4f98a2SYan Zheng } 23955d4f98a2SYan Zheng 23965d4f98a2SYan Zheng if (ret > 0) { 23975d4f98a2SYan Zheng level = ret; 23985d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 23995d4f98a2SYan Zheng path->slots[level]); 24005d4f98a2SYan Zheng replaced = 1; 24015d4f98a2SYan Zheng } 24025d4f98a2SYan Zheng 24035d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 24045d4f98a2SYan Zheng if (ret > 0) 24055d4f98a2SYan Zheng break; 24065d4f98a2SYan Zheng 24075d4f98a2SYan Zheng BUG_ON(level == 0); 24085d4f98a2SYan Zheng /* 24095d4f98a2SYan Zheng * save the merging progress in the drop_progress. 24105d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 24115d4f98a2SYan Zheng */ 24125d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 24135d4f98a2SYan Zheng path->slots[level]); 24145d4f98a2SYan Zheng root_item->drop_level = level; 24155d4f98a2SYan Zheng 24163a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 24179e6a0c52SJosef Bacik trans = NULL; 24185d4f98a2SYan Zheng 24192ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 24205d4f98a2SYan Zheng 24215d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 24225d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 24235d4f98a2SYan Zheng } 24245d4f98a2SYan Zheng 24255d4f98a2SYan Zheng /* 24265d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 24275d4f98a2SYan Zheng * relocated and the block is tree root. 24285d4f98a2SYan Zheng */ 24295d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 24305d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 24315d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 24325d4f98a2SYan Zheng free_extent_buffer(leaf); 24335d4f98a2SYan Zheng if (ret < 0) 24345d4f98a2SYan Zheng err = ret; 24355d4f98a2SYan Zheng out: 24365d4f98a2SYan Zheng btrfs_free_path(path); 24375d4f98a2SYan Zheng 2438d2311e69SQu Wenruo if (err == 0) 2439d2311e69SQu Wenruo insert_dirty_subvol(trans, rc, root); 24405d4f98a2SYan Zheng 24419e6a0c52SJosef Bacik if (trans) 24423a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 24435d4f98a2SYan Zheng 24442ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 24455d4f98a2SYan Zheng 24465d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 24475d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 24485d4f98a2SYan Zheng 24495d4f98a2SYan Zheng return err; 24505d4f98a2SYan Zheng } 24515d4f98a2SYan Zheng 24523fd0a558SYan, Zheng static noinline_for_stack 24533fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 24545d4f98a2SYan Zheng { 24553fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 24560b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 24573fd0a558SYan, Zheng struct btrfs_root *reloc_root; 24585d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 24593fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 24603fd0a558SYan, Zheng u64 num_bytes = 0; 24613fd0a558SYan, Zheng int ret; 24623fd0a558SYan, Zheng 24630b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 24640b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 24653fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 24660b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 24677585717fSChris Mason 24683fd0a558SYan, Zheng again: 24693fd0a558SYan, Zheng if (!err) { 24703fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 247108e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 247208e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 24733fd0a558SYan, Zheng if (ret) 24743fd0a558SYan, Zheng err = ret; 24753fd0a558SYan, Zheng } 24763fd0a558SYan, Zheng 24777a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 24783612b495STsutomu Itoh if (IS_ERR(trans)) { 24793612b495STsutomu Itoh if (!err) 24802ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 2481*63f018beSNikolay Borisov num_bytes, NULL); 24823612b495STsutomu Itoh return PTR_ERR(trans); 24833612b495STsutomu Itoh } 24843fd0a558SYan, Zheng 24853fd0a558SYan, Zheng if (!err) { 24863fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 24873a45bb20SJeff Mahoney btrfs_end_transaction(trans); 24882ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 2489*63f018beSNikolay Borisov num_bytes, NULL); 24903fd0a558SYan, Zheng goto again; 24913fd0a558SYan, Zheng } 24923fd0a558SYan, Zheng } 24933fd0a558SYan, Zheng 24943fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 24953fd0a558SYan, Zheng 24963fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 24973fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 24983fd0a558SYan, Zheng struct btrfs_root, root_list); 24993fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 25003fd0a558SYan, Zheng 25010b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25023fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 25033fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 25043fd0a558SYan, Zheng 25053fd0a558SYan, Zheng /* 25063fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 25073fd0a558SYan, Zheng * knows it should resumes merging 25083fd0a558SYan, Zheng */ 25093fd0a558SYan, Zheng if (!err) 25103fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 25113fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 25123fd0a558SYan, Zheng 25133fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 251400246528SJosef Bacik btrfs_put_root(root); 25153fd0a558SYan, Zheng } 25163fd0a558SYan, Zheng 25173fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 25183fd0a558SYan, Zheng 25193fd0a558SYan, Zheng if (!err) 25203a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 25213fd0a558SYan, Zheng else 25223a45bb20SJeff Mahoney btrfs_end_transaction(trans); 25233fd0a558SYan, Zheng return err; 25243fd0a558SYan, Zheng } 25253fd0a558SYan, Zheng 25263fd0a558SYan, Zheng static noinline_for_stack 2527aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2528aca1bba6SLiu Bo { 2529aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2530aca1bba6SLiu Bo 2531aca1bba6SLiu Bo while (!list_empty(list)) { 2532aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2533aca1bba6SLiu Bo root_list); 2534bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 25356bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 25366bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 25376bdf131fSJosef Bacik reloc_root->node = NULL; 25386bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2539aca1bba6SLiu Bo } 2540aca1bba6SLiu Bo } 2541aca1bba6SLiu Bo 2542aca1bba6SLiu Bo static noinline_for_stack 254394404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 25443fd0a558SYan, Zheng { 25450b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 25465d4f98a2SYan Zheng struct btrfs_root *root; 25475d4f98a2SYan Zheng struct btrfs_root *reloc_root; 25483fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 25493fd0a558SYan, Zheng int found = 0; 2550aca1bba6SLiu Bo int ret = 0; 25513fd0a558SYan, Zheng again: 25523fd0a558SYan, Zheng root = rc->extent_root; 25537585717fSChris Mason 25547585717fSChris Mason /* 25557585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 25567585717fSChris Mason * we have to make sure nobody is in the middle of 25577585717fSChris Mason * adding their roots to the list while we are 25587585717fSChris Mason * doing this splice 25597585717fSChris Mason */ 25600b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 25613fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 25620b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 25635d4f98a2SYan Zheng 25643fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 25653fd0a558SYan, Zheng found = 1; 25663fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 25673fd0a558SYan, Zheng struct btrfs_root, root_list); 25685d4f98a2SYan Zheng 25695d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 25700b246afaSJeff Mahoney root = read_fs_root(fs_info, 25715d4f98a2SYan Zheng reloc_root->root_key.offset); 25725d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25735d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25745d4f98a2SYan Zheng 25753fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 257600246528SJosef Bacik btrfs_put_root(root); 2577b37b39cdSJosef Bacik if (ret) { 257825e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 257925e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 258025e293c2SWang Shilong &reloc_roots); 2581aca1bba6SLiu Bo goto out; 2582b37b39cdSJosef Bacik } 25833fd0a558SYan, Zheng } else { 25843fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 258530d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 258630d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 258730d40577SQu Wenruo &rc->dirty_subvol_roots); 25883fd0a558SYan, Zheng } 25895d4f98a2SYan Zheng } 25905d4f98a2SYan Zheng 25913fd0a558SYan, Zheng if (found) { 25923fd0a558SYan, Zheng found = 0; 25933fd0a558SYan, Zheng goto again; 25945d4f98a2SYan Zheng } 2595aca1bba6SLiu Bo out: 2596aca1bba6SLiu Bo if (ret) { 25970b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2598aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2599aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2600467bb1d2SWang Shilong 2601467bb1d2SWang Shilong /* new reloc root may be added */ 26020b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2603467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 26040b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2605467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2606467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2607aca1bba6SLiu Bo } 2608aca1bba6SLiu Bo 26095d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 26105d4f98a2SYan Zheng } 26115d4f98a2SYan Zheng 26125d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 26135d4f98a2SYan Zheng { 26145d4f98a2SYan Zheng struct tree_block *block; 26155d4f98a2SYan Zheng struct rb_node *rb_node; 26165d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 26175d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 26185d4f98a2SYan Zheng rb_erase(rb_node, blocks); 26195d4f98a2SYan Zheng kfree(block); 26205d4f98a2SYan Zheng } 26215d4f98a2SYan Zheng } 26225d4f98a2SYan Zheng 26235d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 26245d4f98a2SYan Zheng struct btrfs_root *reloc_root) 26255d4f98a2SYan Zheng { 26260b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 26275d4f98a2SYan Zheng struct btrfs_root *root; 2628442b1ac5SJosef Bacik int ret; 26295d4f98a2SYan Zheng 26305d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 26315d4f98a2SYan Zheng return 0; 26325d4f98a2SYan Zheng 26330b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 26345d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 26355d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 2636442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 263700246528SJosef Bacik btrfs_put_root(root); 26385d4f98a2SYan Zheng 2639442b1ac5SJosef Bacik return ret; 26405d4f98a2SYan Zheng } 26415d4f98a2SYan Zheng 26423fd0a558SYan, Zheng static noinline_for_stack 26433fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 26443fd0a558SYan, Zheng struct reloc_control *rc, 26455d4f98a2SYan Zheng struct backref_node *node, 2646dc4103f9SWang Shilong struct backref_edge *edges[]) 26475d4f98a2SYan Zheng { 26485d4f98a2SYan Zheng struct backref_node *next; 26495d4f98a2SYan Zheng struct btrfs_root *root; 26503fd0a558SYan, Zheng int index = 0; 26513fd0a558SYan, Zheng 26525d4f98a2SYan Zheng next = node; 26535d4f98a2SYan Zheng while (1) { 26545d4f98a2SYan Zheng cond_resched(); 26555d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 26565d4f98a2SYan Zheng root = next->root; 26573fd0a558SYan, Zheng BUG_ON(!root); 265827cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 26595d4f98a2SYan Zheng 26605d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 26615d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 26625d4f98a2SYan Zheng break; 26635d4f98a2SYan Zheng } 26645d4f98a2SYan Zheng 26655d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 26663fd0a558SYan, Zheng root = root->reloc_root; 26673fd0a558SYan, Zheng 26683fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 26693fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 26703fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 26713fd0a558SYan, Zheng next->new_bytenr = root->node->start; 267200246528SJosef Bacik btrfs_put_root(next->root); 267300246528SJosef Bacik next->root = btrfs_grab_root(root); 26740b530bc5SJosef Bacik ASSERT(next->root); 26753fd0a558SYan, Zheng list_add_tail(&next->list, 26763fd0a558SYan, Zheng &rc->backref_cache.changed); 26773fd0a558SYan, Zheng __mark_block_processed(rc, next); 26785d4f98a2SYan Zheng break; 26795d4f98a2SYan Zheng } 26805d4f98a2SYan Zheng 26813fd0a558SYan, Zheng WARN_ON(1); 26825d4f98a2SYan Zheng root = NULL; 26835d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 26845d4f98a2SYan Zheng if (!next || next->level <= node->level) 26855d4f98a2SYan Zheng break; 26865d4f98a2SYan Zheng } 26873fd0a558SYan, Zheng if (!root) 26883fd0a558SYan, Zheng return NULL; 26895d4f98a2SYan Zheng 26903fd0a558SYan, Zheng next = node; 26913fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 26923fd0a558SYan, Zheng while (1) { 26933fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 26943fd0a558SYan, Zheng if (--index < 0) 26953fd0a558SYan, Zheng break; 26963fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 26973fd0a558SYan, Zheng } 26985d4f98a2SYan Zheng return root; 26995d4f98a2SYan Zheng } 27005d4f98a2SYan Zheng 27013fd0a558SYan, Zheng /* 27023fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 27033fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 27043fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 27053fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 27063fd0a558SYan, Zheng */ 27075d4f98a2SYan Zheng static noinline_for_stack 2708147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 27095d4f98a2SYan Zheng { 27103fd0a558SYan, Zheng struct backref_node *next; 27113fd0a558SYan, Zheng struct btrfs_root *root; 27123fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 27135d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27143fd0a558SYan, Zheng int index = 0; 27153fd0a558SYan, Zheng 27163fd0a558SYan, Zheng next = node; 27173fd0a558SYan, Zheng while (1) { 27183fd0a558SYan, Zheng cond_resched(); 27193fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 27203fd0a558SYan, Zheng root = next->root; 27213fd0a558SYan, Zheng BUG_ON(!root); 27223fd0a558SYan, Zheng 272325985edcSLucas De Marchi /* no other choice for non-references counted tree */ 272427cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 27253fd0a558SYan, Zheng return root; 27263fd0a558SYan, Zheng 27273fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 27283fd0a558SYan, Zheng fs_root = root; 27293fd0a558SYan, Zheng 27303fd0a558SYan, Zheng if (next != node) 27313fd0a558SYan, Zheng return NULL; 27323fd0a558SYan, Zheng 27333fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 27343fd0a558SYan, Zheng if (!next || next->level <= node->level) 27353fd0a558SYan, Zheng break; 27363fd0a558SYan, Zheng } 27373fd0a558SYan, Zheng 27383fd0a558SYan, Zheng if (!fs_root) 27393fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 27403fd0a558SYan, Zheng return fs_root; 27415d4f98a2SYan Zheng } 27425d4f98a2SYan Zheng 27435d4f98a2SYan Zheng static noinline_for_stack 27443fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 27453fd0a558SYan, Zheng struct backref_node *node, int reserve) 27465d4f98a2SYan Zheng { 27470b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 27483fd0a558SYan, Zheng struct backref_node *next = node; 27493fd0a558SYan, Zheng struct backref_edge *edge; 27503fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27513fd0a558SYan, Zheng u64 num_bytes = 0; 27523fd0a558SYan, Zheng int index = 0; 27535d4f98a2SYan Zheng 27543fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 27553fd0a558SYan, Zheng 27563fd0a558SYan, Zheng while (next) { 27573fd0a558SYan, Zheng cond_resched(); 27585d4f98a2SYan Zheng while (1) { 27593fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 27605d4f98a2SYan Zheng break; 27615d4f98a2SYan Zheng 27620b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 27633fd0a558SYan, Zheng 27643fd0a558SYan, Zheng if (list_empty(&next->upper)) 27653fd0a558SYan, Zheng break; 27663fd0a558SYan, Zheng 27673fd0a558SYan, Zheng edge = list_entry(next->upper.next, 27683fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 27693fd0a558SYan, Zheng edges[index++] = edge; 27703fd0a558SYan, Zheng next = edge->node[UPPER]; 27715d4f98a2SYan Zheng } 27723fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 27733fd0a558SYan, Zheng } 27743fd0a558SYan, Zheng return num_bytes; 27753fd0a558SYan, Zheng } 27763fd0a558SYan, Zheng 27773fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 27783fd0a558SYan, Zheng struct reloc_control *rc, 27793fd0a558SYan, Zheng struct backref_node *node) 27803fd0a558SYan, Zheng { 27813fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2782da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 27833fd0a558SYan, Zheng u64 num_bytes; 27843fd0a558SYan, Zheng int ret; 27850647bf56SWang Shilong u64 tmp; 27863fd0a558SYan, Zheng 27873fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 27883fd0a558SYan, Zheng 27893fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 27900647bf56SWang Shilong rc->reserved_bytes += num_bytes; 27918ca17f0fSJosef Bacik 27928ca17f0fSJosef Bacik /* 27938ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 27948ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 27958ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 27968ca17f0fSJosef Bacik */ 27970647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 27988ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 27993fd0a558SYan, Zheng if (ret) { 2800da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 28010647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 28020647bf56SWang Shilong tmp <<= 1; 28030647bf56SWang Shilong /* 28040647bf56SWang Shilong * only one thread can access block_rsv at this point, 28050647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 28060647bf56SWang Shilong * we expand more reservation size here to allow enough 280752042d8eSAndrea Gelmini * space for relocation and we will return earlier in 28080647bf56SWang Shilong * enospc case. 28090647bf56SWang Shilong */ 2810da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 28110647bf56SWang Shilong RELOCATION_RESERVED_NODES; 28128ca17f0fSJosef Bacik return -EAGAIN; 28133fd0a558SYan, Zheng } 28143fd0a558SYan, Zheng 28153fd0a558SYan, Zheng return 0; 28163fd0a558SYan, Zheng } 28173fd0a558SYan, Zheng 28185d4f98a2SYan Zheng /* 28195d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 28205d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 28215d4f98a2SYan Zheng * 28225d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 28235d4f98a2SYan Zheng * in that case this function just updates pointers. 28245d4f98a2SYan Zheng */ 28255d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 28263fd0a558SYan, Zheng struct reloc_control *rc, 28275d4f98a2SYan Zheng struct backref_node *node, 28285d4f98a2SYan Zheng struct btrfs_key *key, 28295d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 28305d4f98a2SYan Zheng { 28312ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 28325d4f98a2SYan Zheng struct backref_node *upper; 28335d4f98a2SYan Zheng struct backref_edge *edge; 28345d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 28355d4f98a2SYan Zheng struct btrfs_root *root; 28365d4f98a2SYan Zheng struct extent_buffer *eb; 28375d4f98a2SYan Zheng u32 blocksize; 28385d4f98a2SYan Zheng u64 bytenr; 28395d4f98a2SYan Zheng u64 generation; 28405d4f98a2SYan Zheng int slot; 28415d4f98a2SYan Zheng int ret; 28425d4f98a2SYan Zheng int err = 0; 28435d4f98a2SYan Zheng 28445d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 28455d4f98a2SYan Zheng 28465d4f98a2SYan Zheng path->lowest_level = node->level + 1; 28473fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 28485d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 2849581c1760SQu Wenruo struct btrfs_key first_key; 285082fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2851581c1760SQu Wenruo 28525d4f98a2SYan Zheng cond_resched(); 28535d4f98a2SYan Zheng 28545d4f98a2SYan Zheng upper = edge->node[UPPER]; 2855dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 28563fd0a558SYan, Zheng BUG_ON(!root); 28575d4f98a2SYan Zheng 28583fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 28593fd0a558SYan, Zheng if (!lowest) { 28603fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 28613fd0a558SYan, Zheng upper->level, &slot); 2862cbca7d59SFilipe Manana if (ret < 0) { 2863cbca7d59SFilipe Manana err = ret; 2864cbca7d59SFilipe Manana goto next; 2865cbca7d59SFilipe Manana } 28663fd0a558SYan, Zheng BUG_ON(ret); 28673fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 28683fd0a558SYan, Zheng if (node->eb->start == bytenr) 28693fd0a558SYan, Zheng goto next; 28703fd0a558SYan, Zheng } 28715d4f98a2SYan Zheng drop_node_buffer(upper); 28723fd0a558SYan, Zheng } 28735d4f98a2SYan Zheng 28745d4f98a2SYan Zheng if (!upper->eb) { 28755d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 28763561b9dbSLiu Bo if (ret) { 28773561b9dbSLiu Bo if (ret < 0) 28785d4f98a2SYan Zheng err = ret; 28793561b9dbSLiu Bo else 28803561b9dbSLiu Bo err = -ENOENT; 28813561b9dbSLiu Bo 28823561b9dbSLiu Bo btrfs_release_path(path); 28835d4f98a2SYan Zheng break; 28845d4f98a2SYan Zheng } 28855d4f98a2SYan Zheng 28863fd0a558SYan, Zheng if (!upper->eb) { 28873fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 28883fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 28893fd0a558SYan, Zheng } else { 28903fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 28913fd0a558SYan, Zheng } 28923fd0a558SYan, Zheng 28933fd0a558SYan, Zheng upper->locked = 1; 28943fd0a558SYan, Zheng path->locks[upper->level] = 0; 28953fd0a558SYan, Zheng 28965d4f98a2SYan Zheng slot = path->slots[upper->level]; 2897b3b4aa74SDavid Sterba btrfs_release_path(path); 28985d4f98a2SYan Zheng } else { 28995d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 29005d4f98a2SYan Zheng &slot); 2901cbca7d59SFilipe Manana if (ret < 0) { 2902cbca7d59SFilipe Manana err = ret; 2903cbca7d59SFilipe Manana goto next; 2904cbca7d59SFilipe Manana } 29055d4f98a2SYan Zheng BUG_ON(ret); 29065d4f98a2SYan Zheng } 29075d4f98a2SYan Zheng 29085d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 29093fd0a558SYan, Zheng if (lowest) { 29104547f4d8SLiu Bo if (bytenr != node->bytenr) { 29114547f4d8SLiu Bo btrfs_err(root->fs_info, 29124547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 29134547f4d8SLiu Bo bytenr, node->bytenr, slot, 29144547f4d8SLiu Bo upper->eb->start); 29154547f4d8SLiu Bo err = -EIO; 29164547f4d8SLiu Bo goto next; 29174547f4d8SLiu Bo } 29185d4f98a2SYan Zheng } else { 29193fd0a558SYan, Zheng if (node->eb->start == bytenr) 29203fd0a558SYan, Zheng goto next; 29215d4f98a2SYan Zheng } 29225d4f98a2SYan Zheng 2923da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 29245d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2925581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 2926581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 2927581c1760SQu Wenruo upper->level - 1, &first_key); 292864c043deSLiu Bo if (IS_ERR(eb)) { 292964c043deSLiu Bo err = PTR_ERR(eb); 293064c043deSLiu Bo goto next; 293164c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2932416bc658SJosef Bacik free_extent_buffer(eb); 293397d9a8a4STsutomu Itoh err = -EIO; 293497d9a8a4STsutomu Itoh goto next; 293597d9a8a4STsutomu Itoh } 29365d4f98a2SYan Zheng btrfs_tree_lock(eb); 29378bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 29385d4f98a2SYan Zheng 29395d4f98a2SYan Zheng if (!node->eb) { 29405d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 29415d4f98a2SYan Zheng slot, &eb); 29423fd0a558SYan, Zheng btrfs_tree_unlock(eb); 29433fd0a558SYan, Zheng free_extent_buffer(eb); 29445d4f98a2SYan Zheng if (ret < 0) { 29455d4f98a2SYan Zheng err = ret; 29463fd0a558SYan, Zheng goto next; 29475d4f98a2SYan Zheng } 29483fd0a558SYan, Zheng BUG_ON(node->eb != eb); 29495d4f98a2SYan Zheng } else { 29505d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 29515d4f98a2SYan Zheng node->eb->start); 29525d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 29535d4f98a2SYan Zheng trans->transid); 29545d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 29555d4f98a2SYan Zheng 295682fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 29575d4f98a2SYan Zheng node->eb->start, blocksize, 295882fa113fSQu Wenruo upper->eb->start); 295982fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 296082fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 296182fa113fSQu Wenruo btrfs_header_owner(upper->eb)); 296282fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 29635d4f98a2SYan Zheng BUG_ON(ret); 29645d4f98a2SYan Zheng 29655d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 29665d4f98a2SYan Zheng BUG_ON(ret); 29675d4f98a2SYan Zheng } 29683fd0a558SYan, Zheng next: 29693fd0a558SYan, Zheng if (!upper->pending) 29703fd0a558SYan, Zheng drop_node_buffer(upper); 29713fd0a558SYan, Zheng else 29723fd0a558SYan, Zheng unlock_node_buffer(upper); 29733fd0a558SYan, Zheng if (err) 29743fd0a558SYan, Zheng break; 29755d4f98a2SYan Zheng } 29763fd0a558SYan, Zheng 29773fd0a558SYan, Zheng if (!err && node->pending) { 29783fd0a558SYan, Zheng drop_node_buffer(node); 29793fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 29803fd0a558SYan, Zheng node->pending = 0; 29815d4f98a2SYan Zheng } 29823fd0a558SYan, Zheng 29835d4f98a2SYan Zheng path->lowest_level = 0; 29843fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 29855d4f98a2SYan Zheng return err; 29865d4f98a2SYan Zheng } 29875d4f98a2SYan Zheng 29885d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 29893fd0a558SYan, Zheng struct reloc_control *rc, 29905d4f98a2SYan Zheng struct backref_node *node, 29915d4f98a2SYan Zheng struct btrfs_path *path) 29925d4f98a2SYan Zheng { 29935d4f98a2SYan Zheng struct btrfs_key key; 29945d4f98a2SYan Zheng 29955d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 29963fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 29975d4f98a2SYan Zheng } 29985d4f98a2SYan Zheng 29995d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 30003fd0a558SYan, Zheng struct reloc_control *rc, 30013fd0a558SYan, Zheng struct btrfs_path *path, int err) 30025d4f98a2SYan Zheng { 30033fd0a558SYan, Zheng LIST_HEAD(list); 30043fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 30055d4f98a2SYan Zheng struct backref_node *node; 30065d4f98a2SYan Zheng int level; 30075d4f98a2SYan Zheng int ret; 30085d4f98a2SYan Zheng 30095d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 30105d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 30115d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 30123fd0a558SYan, Zheng struct backref_node, list); 30133fd0a558SYan, Zheng list_move_tail(&node->list, &list); 30143fd0a558SYan, Zheng BUG_ON(!node->pending); 30155d4f98a2SYan Zheng 30163fd0a558SYan, Zheng if (!err) { 30173fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 30185d4f98a2SYan Zheng if (ret < 0) 30195d4f98a2SYan Zheng err = ret; 30205d4f98a2SYan Zheng } 30215d4f98a2SYan Zheng } 30223fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 30233fd0a558SYan, Zheng } 30245d4f98a2SYan Zheng return err; 30255d4f98a2SYan Zheng } 30265d4f98a2SYan Zheng 30275d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 30283fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 30293fd0a558SYan, Zheng { 30303fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 3031ceeb0ae7SDavid Sterba EXTENT_DIRTY); 30323fd0a558SYan, Zheng } 30333fd0a558SYan, Zheng 30343fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 30355d4f98a2SYan Zheng struct backref_node *node) 30365d4f98a2SYan Zheng { 30375d4f98a2SYan Zheng u32 blocksize; 30385d4f98a2SYan Zheng if (node->level == 0 || 30395d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 3040da17066cSJeff Mahoney blocksize = rc->extent_root->fs_info->nodesize; 30413fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 30425d4f98a2SYan Zheng } 30435d4f98a2SYan Zheng node->processed = 1; 30445d4f98a2SYan Zheng } 30455d4f98a2SYan Zheng 30465d4f98a2SYan Zheng /* 30475d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 30485d4f98a2SYan Zheng * as processed. 30495d4f98a2SYan Zheng */ 30505d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 30515d4f98a2SYan Zheng struct backref_node *node) 30525d4f98a2SYan Zheng { 30535d4f98a2SYan Zheng struct backref_node *next = node; 30545d4f98a2SYan Zheng struct backref_edge *edge; 30555d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 30565d4f98a2SYan Zheng int index = 0; 30575d4f98a2SYan Zheng 30585d4f98a2SYan Zheng while (next) { 30595d4f98a2SYan Zheng cond_resched(); 30605d4f98a2SYan Zheng while (1) { 30615d4f98a2SYan Zheng if (next->processed) 30625d4f98a2SYan Zheng break; 30635d4f98a2SYan Zheng 30643fd0a558SYan, Zheng __mark_block_processed(rc, next); 30655d4f98a2SYan Zheng 30665d4f98a2SYan Zheng if (list_empty(&next->upper)) 30675d4f98a2SYan Zheng break; 30685d4f98a2SYan Zheng 30695d4f98a2SYan Zheng edge = list_entry(next->upper.next, 30705d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 30715d4f98a2SYan Zheng edges[index++] = edge; 30725d4f98a2SYan Zheng next = edge->node[UPPER]; 30735d4f98a2SYan Zheng } 30745d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 30755d4f98a2SYan Zheng } 30765d4f98a2SYan Zheng } 30775d4f98a2SYan Zheng 30787476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 30795d4f98a2SYan Zheng { 3080da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 30817476dfdaSDavid Sterba 30825d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 30839655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 30845d4f98a2SYan Zheng return 1; 30855d4f98a2SYan Zheng return 0; 30865d4f98a2SYan Zheng } 30875d4f98a2SYan Zheng 30882ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 30895d4f98a2SYan Zheng struct tree_block *block) 30905d4f98a2SYan Zheng { 30915d4f98a2SYan Zheng struct extent_buffer *eb; 30925d4f98a2SYan Zheng 3093581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 3094581c1760SQu Wenruo block->level, NULL); 309564c043deSLiu Bo if (IS_ERR(eb)) { 309664c043deSLiu Bo return PTR_ERR(eb); 309764c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 3098416bc658SJosef Bacik free_extent_buffer(eb); 3099416bc658SJosef Bacik return -EIO; 3100416bc658SJosef Bacik } 31015d4f98a2SYan Zheng if (block->level == 0) 31025d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 31035d4f98a2SYan Zheng else 31045d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 31055d4f98a2SYan Zheng free_extent_buffer(eb); 31065d4f98a2SYan Zheng block->key_ready = 1; 31075d4f98a2SYan Zheng return 0; 31085d4f98a2SYan Zheng } 31095d4f98a2SYan Zheng 31105d4f98a2SYan Zheng /* 31115d4f98a2SYan Zheng * helper function to relocate a tree block 31125d4f98a2SYan Zheng */ 31135d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 31145d4f98a2SYan Zheng struct reloc_control *rc, 31155d4f98a2SYan Zheng struct backref_node *node, 31165d4f98a2SYan Zheng struct btrfs_key *key, 31175d4f98a2SYan Zheng struct btrfs_path *path) 31185d4f98a2SYan Zheng { 31195d4f98a2SYan Zheng struct btrfs_root *root; 31203fd0a558SYan, Zheng int ret = 0; 31215d4f98a2SYan Zheng 31223fd0a558SYan, Zheng if (!node) 31235d4f98a2SYan Zheng return 0; 31243fd0a558SYan, Zheng 31253fd0a558SYan, Zheng BUG_ON(node->processed); 3126147d256eSZhaolei root = select_one_root(node); 31273fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 31283fd0a558SYan, Zheng update_processed_blocks(rc, node); 31293fd0a558SYan, Zheng goto out; 31305d4f98a2SYan Zheng } 31315d4f98a2SYan Zheng 313227cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 31333fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 31343fd0a558SYan, Zheng if (ret) 31355d4f98a2SYan Zheng goto out; 31365d4f98a2SYan Zheng } 31373fd0a558SYan, Zheng 31383fd0a558SYan, Zheng if (root) { 313927cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 31403fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 31413fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 31423fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 31433fd0a558SYan, Zheng root = root->reloc_root; 31443fd0a558SYan, Zheng node->new_bytenr = root->node->start; 314500246528SJosef Bacik btrfs_put_root(node->root); 314600246528SJosef Bacik node->root = btrfs_grab_root(root); 31470b530bc5SJosef Bacik ASSERT(node->root); 31483fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 31493fd0a558SYan, Zheng } else { 31505d4f98a2SYan Zheng path->lowest_level = node->level; 31515d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3152b3b4aa74SDavid Sterba btrfs_release_path(path); 31533fd0a558SYan, Zheng if (ret > 0) 31545d4f98a2SYan Zheng ret = 0; 31553fd0a558SYan, Zheng } 31563fd0a558SYan, Zheng if (!ret) 31573fd0a558SYan, Zheng update_processed_blocks(rc, node); 31583fd0a558SYan, Zheng } else { 31593fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 31603fd0a558SYan, Zheng } 31615d4f98a2SYan Zheng out: 31620647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 31633fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 31645d4f98a2SYan Zheng return ret; 31655d4f98a2SYan Zheng } 31665d4f98a2SYan Zheng 31675d4f98a2SYan Zheng /* 31685d4f98a2SYan Zheng * relocate a list of blocks 31695d4f98a2SYan Zheng */ 31705d4f98a2SYan Zheng static noinline_for_stack 31715d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 31725d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 31735d4f98a2SYan Zheng { 31742ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 31755d4f98a2SYan Zheng struct backref_node *node; 31765d4f98a2SYan Zheng struct btrfs_path *path; 31775d4f98a2SYan Zheng struct tree_block *block; 317898ff7b94SQu Wenruo struct tree_block *next; 31795d4f98a2SYan Zheng int ret; 31805d4f98a2SYan Zheng int err = 0; 31815d4f98a2SYan Zheng 31825d4f98a2SYan Zheng path = btrfs_alloc_path(); 3183e1a12670SLiu Bo if (!path) { 3184e1a12670SLiu Bo err = -ENOMEM; 318534c2b290SDavid Sterba goto out_free_blocks; 3186e1a12670SLiu Bo } 31875d4f98a2SYan Zheng 318898ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 318998ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 31905d4f98a2SYan Zheng if (!block->key_ready) 31912ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 31925d4f98a2SYan Zheng } 31935d4f98a2SYan Zheng 319498ff7b94SQu Wenruo /* Get first keys */ 319598ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 319634c2b290SDavid Sterba if (!block->key_ready) { 31972ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 319834c2b290SDavid Sterba if (err) 319934c2b290SDavid Sterba goto out_free_path; 320034c2b290SDavid Sterba } 32015d4f98a2SYan Zheng } 32025d4f98a2SYan Zheng 320398ff7b94SQu Wenruo /* Do tree relocation */ 320498ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 32053fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 32065d4f98a2SYan Zheng block->level, block->bytenr); 32075d4f98a2SYan Zheng if (IS_ERR(node)) { 32085d4f98a2SYan Zheng err = PTR_ERR(node); 32095d4f98a2SYan Zheng goto out; 32105d4f98a2SYan Zheng } 32115d4f98a2SYan Zheng 32125d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 32135d4f98a2SYan Zheng path); 32145d4f98a2SYan Zheng if (ret < 0) { 321598ff7b94SQu Wenruo if (ret != -EAGAIN || &block->rb_node == rb_first(blocks)) 32165d4f98a2SYan Zheng err = ret; 32175d4f98a2SYan Zheng goto out; 32185d4f98a2SYan Zheng } 32195d4f98a2SYan Zheng } 32205d4f98a2SYan Zheng out: 32213fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 32225d4f98a2SYan Zheng 322334c2b290SDavid Sterba out_free_path: 32245d4f98a2SYan Zheng btrfs_free_path(path); 322534c2b290SDavid Sterba out_free_blocks: 3226e1a12670SLiu Bo free_block_list(blocks); 32275d4f98a2SYan Zheng return err; 32285d4f98a2SYan Zheng } 32295d4f98a2SYan Zheng 32305d4f98a2SYan Zheng static noinline_for_stack 3231efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3232efa56464SYan, Zheng struct file_extent_cluster *cluster) 3233efa56464SYan, Zheng { 3234efa56464SYan, Zheng u64 alloc_hint = 0; 3235efa56464SYan, Zheng u64 start; 3236efa56464SYan, Zheng u64 end; 3237efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3238efa56464SYan, Zheng u64 num_bytes; 3239efa56464SYan, Zheng int nr = 0; 3240efa56464SYan, Zheng int ret = 0; 3241dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3242dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 324318513091SWang Xiaoguang u64 cur_offset; 3244364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3245efa56464SYan, Zheng 3246efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 32475955102cSAl Viro inode_lock(inode); 3248efa56464SYan, Zheng 3249364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3250dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3251efa56464SYan, Zheng if (ret) 3252efa56464SYan, Zheng goto out; 3253efa56464SYan, Zheng 325418513091SWang Xiaoguang cur_offset = prealloc_start; 3255efa56464SYan, Zheng while (nr < cluster->nr) { 3256efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3257efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3258efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3259efa56464SYan, Zheng else 3260efa56464SYan, Zheng end = cluster->end - offset; 3261efa56464SYan, Zheng 3262d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3263efa56464SYan, Zheng num_bytes = end + 1 - start; 326418513091SWang Xiaoguang if (cur_offset < start) 3265bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3266bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3267efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3268efa56464SYan, Zheng num_bytes, num_bytes, 3269efa56464SYan, Zheng end + 1, &alloc_hint); 327018513091SWang Xiaoguang cur_offset = end + 1; 3271d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3272efa56464SYan, Zheng if (ret) 3273efa56464SYan, Zheng break; 3274efa56464SYan, Zheng nr++; 3275efa56464SYan, Zheng } 327618513091SWang Xiaoguang if (cur_offset < prealloc_end) 3277bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3278bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3279efa56464SYan, Zheng out: 32805955102cSAl Viro inode_unlock(inode); 3281364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3282efa56464SYan, Zheng return ret; 3283efa56464SYan, Zheng } 3284efa56464SYan, Zheng 3285efa56464SYan, Zheng static noinline_for_stack 32860257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 32870257bb82SYan, Zheng u64 block_start) 32885d4f98a2SYan Zheng { 32895d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 32905d4f98a2SYan Zheng struct extent_map *em; 32910257bb82SYan, Zheng int ret = 0; 32925d4f98a2SYan Zheng 3293172ddd60SDavid Sterba em = alloc_extent_map(); 32940257bb82SYan, Zheng if (!em) 32950257bb82SYan, Zheng return -ENOMEM; 32960257bb82SYan, Zheng 32975d4f98a2SYan Zheng em->start = start; 32980257bb82SYan, Zheng em->len = end + 1 - start; 32990257bb82SYan, Zheng em->block_len = em->len; 33000257bb82SYan, Zheng em->block_start = block_start; 33015d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 33025d4f98a2SYan Zheng 3303d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 33045d4f98a2SYan Zheng while (1) { 3305890871beSChris Mason write_lock(&em_tree->lock); 330609a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3307890871beSChris Mason write_unlock(&em_tree->lock); 33085d4f98a2SYan Zheng if (ret != -EEXIST) { 33095d4f98a2SYan Zheng free_extent_map(em); 33105d4f98a2SYan Zheng break; 33115d4f98a2SYan Zheng } 3312dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 33135d4f98a2SYan Zheng } 3314d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 33150257bb82SYan, Zheng return ret; 33160257bb82SYan, Zheng } 33175d4f98a2SYan Zheng 3318726a3421SQu Wenruo /* 3319726a3421SQu Wenruo * Allow error injection to test balance cancellation 3320726a3421SQu Wenruo */ 3321726a3421SQu Wenruo int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 3322726a3421SQu Wenruo { 3323726a3421SQu Wenruo return atomic_read(&fs_info->balance_cancel_req); 3324726a3421SQu Wenruo } 3325726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 3326726a3421SQu Wenruo 33270257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 33280257bb82SYan, Zheng struct file_extent_cluster *cluster) 33290257bb82SYan, Zheng { 33302ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 33310257bb82SYan, Zheng u64 page_start; 33320257bb82SYan, Zheng u64 page_end; 33330257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 33340257bb82SYan, Zheng unsigned long index; 33350257bb82SYan, Zheng unsigned long last_index; 33360257bb82SYan, Zheng struct page *page; 33370257bb82SYan, Zheng struct file_ra_state *ra; 33383b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 33390257bb82SYan, Zheng int nr = 0; 33400257bb82SYan, Zheng int ret = 0; 33410257bb82SYan, Zheng 33420257bb82SYan, Zheng if (!cluster->nr) 33430257bb82SYan, Zheng return 0; 33440257bb82SYan, Zheng 33450257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 33460257bb82SYan, Zheng if (!ra) 33470257bb82SYan, Zheng return -ENOMEM; 33480257bb82SYan, Zheng 3349efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 33500257bb82SYan, Zheng if (ret) 3351efa56464SYan, Zheng goto out; 33520257bb82SYan, Zheng 33530257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 33540257bb82SYan, Zheng 3355efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3356efa56464SYan, Zheng cluster->end - offset, cluster->start); 3357efa56464SYan, Zheng if (ret) 3358efa56464SYan, Zheng goto out; 3359efa56464SYan, Zheng 336009cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 336109cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 33620257bb82SYan, Zheng while (index <= last_index) { 33639f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 33649f3db423SNikolay Borisov PAGE_SIZE); 3365efa56464SYan, Zheng if (ret) 3366efa56464SYan, Zheng goto out; 3367efa56464SYan, Zheng 33680257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 33690257bb82SYan, Zheng if (!page) { 33700257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 33710257bb82SYan, Zheng ra, NULL, index, 33720257bb82SYan, Zheng last_index + 1 - index); 3373a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 33743b16a4e3SJosef Bacik mask); 33750257bb82SYan, Zheng if (!page) { 3376691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 337743b18595SQu Wenruo PAGE_SIZE, true); 337844db1216SFilipe Manana btrfs_delalloc_release_extents(BTRFS_I(inode), 33798702ba93SQu Wenruo PAGE_SIZE); 33800257bb82SYan, Zheng ret = -ENOMEM; 3381efa56464SYan, Zheng goto out; 33820257bb82SYan, Zheng } 33830257bb82SYan, Zheng } 33840257bb82SYan, Zheng 33850257bb82SYan, Zheng if (PageReadahead(page)) { 33860257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 33870257bb82SYan, Zheng ra, NULL, page, index, 33880257bb82SYan, Zheng last_index + 1 - index); 33890257bb82SYan, Zheng } 33900257bb82SYan, Zheng 33910257bb82SYan, Zheng if (!PageUptodate(page)) { 33920257bb82SYan, Zheng btrfs_readpage(NULL, page); 33930257bb82SYan, Zheng lock_page(page); 33940257bb82SYan, Zheng if (!PageUptodate(page)) { 33950257bb82SYan, Zheng unlock_page(page); 339609cbfeafSKirill A. Shutemov put_page(page); 3397691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 339843b18595SQu Wenruo PAGE_SIZE, true); 33998b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 34008702ba93SQu Wenruo PAGE_SIZE); 34010257bb82SYan, Zheng ret = -EIO; 3402efa56464SYan, Zheng goto out; 34030257bb82SYan, Zheng } 34040257bb82SYan, Zheng } 34050257bb82SYan, Zheng 34064eee4fa4SMiao Xie page_start = page_offset(page); 340709cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 34080257bb82SYan, Zheng 3409d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 34100257bb82SYan, Zheng 34110257bb82SYan, Zheng set_page_extent_mapped(page); 34120257bb82SYan, Zheng 34130257bb82SYan, Zheng if (nr < cluster->nr && 34140257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 34150257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 34160257bb82SYan, Zheng page_start, page_end, 3417ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 34180257bb82SYan, Zheng nr++; 34190257bb82SYan, Zheng } 34200257bb82SYan, Zheng 3421765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3422330a5827SNikolay Borisov NULL); 3423765f3cebSNikolay Borisov if (ret) { 3424765f3cebSNikolay Borisov unlock_page(page); 3425765f3cebSNikolay Borisov put_page(page); 3426765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 342743b18595SQu Wenruo PAGE_SIZE, true); 3428765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 34298702ba93SQu Wenruo PAGE_SIZE); 3430765f3cebSNikolay Borisov 3431765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3432765f3cebSNikolay Borisov page_start, page_end, 3433765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3434765f3cebSNikolay Borisov goto out; 3435765f3cebSNikolay Borisov 3436765f3cebSNikolay Borisov } 34370257bb82SYan, Zheng set_page_dirty(page); 34380257bb82SYan, Zheng 34390257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3440d0082371SJeff Mahoney page_start, page_end); 34410257bb82SYan, Zheng unlock_page(page); 344209cbfeafSKirill A. Shutemov put_page(page); 34430257bb82SYan, Zheng 34440257bb82SYan, Zheng index++; 34458702ba93SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); 3446efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 34472ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 34487f913c7cSQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 34497f913c7cSQu Wenruo ret = -ECANCELED; 34507f913c7cSQu Wenruo goto out; 34517f913c7cSQu Wenruo } 34520257bb82SYan, Zheng } 34530257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3454efa56464SYan, Zheng out: 34550257bb82SYan, Zheng kfree(ra); 34560257bb82SYan, Zheng return ret; 34570257bb82SYan, Zheng } 34580257bb82SYan, Zheng 34590257bb82SYan, Zheng static noinline_for_stack 34600257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 34610257bb82SYan, Zheng struct file_extent_cluster *cluster) 34620257bb82SYan, Zheng { 34630257bb82SYan, Zheng int ret; 34640257bb82SYan, Zheng 34650257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 34660257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 34670257bb82SYan, Zheng if (ret) 34680257bb82SYan, Zheng return ret; 34690257bb82SYan, Zheng cluster->nr = 0; 34700257bb82SYan, Zheng } 34710257bb82SYan, Zheng 34720257bb82SYan, Zheng if (!cluster->nr) 34730257bb82SYan, Zheng cluster->start = extent_key->objectid; 34740257bb82SYan, Zheng else 34750257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 34760257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 34770257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 34780257bb82SYan, Zheng cluster->nr++; 34790257bb82SYan, Zheng 34800257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 34810257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 34820257bb82SYan, Zheng if (ret) 34830257bb82SYan, Zheng return ret; 34840257bb82SYan, Zheng cluster->nr = 0; 34850257bb82SYan, Zheng } 34860257bb82SYan, Zheng return 0; 34875d4f98a2SYan Zheng } 34885d4f98a2SYan Zheng 34895d4f98a2SYan Zheng /* 34905d4f98a2SYan Zheng * helper to add a tree block to the list. 34915d4f98a2SYan Zheng * the major work is getting the generation and level of the block 34925d4f98a2SYan Zheng */ 34935d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 34945d4f98a2SYan Zheng struct btrfs_key *extent_key, 34955d4f98a2SYan Zheng struct btrfs_path *path, 34965d4f98a2SYan Zheng struct rb_root *blocks) 34975d4f98a2SYan Zheng { 34985d4f98a2SYan Zheng struct extent_buffer *eb; 34995d4f98a2SYan Zheng struct btrfs_extent_item *ei; 35005d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 35015d4f98a2SYan Zheng struct tree_block *block; 35025d4f98a2SYan Zheng struct rb_node *rb_node; 35035d4f98a2SYan Zheng u32 item_size; 35045d4f98a2SYan Zheng int level = -1; 35057fdf4b60SWang Shilong u64 generation; 35065d4f98a2SYan Zheng 35075d4f98a2SYan Zheng eb = path->nodes[0]; 35085d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 35095d4f98a2SYan Zheng 35103173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 35113173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 35125d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 35135d4f98a2SYan Zheng struct btrfs_extent_item); 35143173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 35155d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 35165d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 35175d4f98a2SYan Zheng } else { 35183173a18fSJosef Bacik level = (int)extent_key->offset; 35193173a18fSJosef Bacik } 35203173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 35216d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3522ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3523ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3524ba3c2b19SNikolay Borisov return -EINVAL; 35253173a18fSJosef Bacik } else { 35265d4f98a2SYan Zheng BUG(); 35275d4f98a2SYan Zheng } 35285d4f98a2SYan Zheng 3529b3b4aa74SDavid Sterba btrfs_release_path(path); 35305d4f98a2SYan Zheng 35315d4f98a2SYan Zheng BUG_ON(level == -1); 35325d4f98a2SYan Zheng 35335d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 35345d4f98a2SYan Zheng if (!block) 35355d4f98a2SYan Zheng return -ENOMEM; 35365d4f98a2SYan Zheng 35375d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3538da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 35395d4f98a2SYan Zheng block->key.offset = generation; 35405d4f98a2SYan Zheng block->level = level; 35415d4f98a2SYan Zheng block->key_ready = 0; 35425d4f98a2SYan Zheng 35435d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 354443c04fb1SJeff Mahoney if (rb_node) 354543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 35465d4f98a2SYan Zheng 35475d4f98a2SYan Zheng return 0; 35485d4f98a2SYan Zheng } 35495d4f98a2SYan Zheng 35505d4f98a2SYan Zheng /* 35515d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 35525d4f98a2SYan Zheng */ 35535d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 35545d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 35555d4f98a2SYan Zheng struct rb_root *blocks) 35565d4f98a2SYan Zheng { 35570b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35585d4f98a2SYan Zheng struct btrfs_path *path; 35595d4f98a2SYan Zheng struct btrfs_key key; 35605d4f98a2SYan Zheng int ret; 35610b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 35625d4f98a2SYan Zheng 35637476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 35645d4f98a2SYan Zheng return 0; 35655d4f98a2SYan Zheng 35665d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 35675d4f98a2SYan Zheng return 0; 35685d4f98a2SYan Zheng 35695d4f98a2SYan Zheng path = btrfs_alloc_path(); 35705d4f98a2SYan Zheng if (!path) 35715d4f98a2SYan Zheng return -ENOMEM; 3572aee68ee5SJosef Bacik again: 35735d4f98a2SYan Zheng key.objectid = bytenr; 3574aee68ee5SJosef Bacik if (skinny) { 3575aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3576aee68ee5SJosef Bacik key.offset = (u64)-1; 3577aee68ee5SJosef Bacik } else { 35785d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 35795d4f98a2SYan Zheng key.offset = blocksize; 3580aee68ee5SJosef Bacik } 35815d4f98a2SYan Zheng 35825d4f98a2SYan Zheng path->search_commit_root = 1; 35835d4f98a2SYan Zheng path->skip_locking = 1; 35845d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 35855d4f98a2SYan Zheng if (ret < 0) 35865d4f98a2SYan Zheng goto out; 35875d4f98a2SYan Zheng 3588aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3589aee68ee5SJosef Bacik if (path->slots[0]) { 3590aee68ee5SJosef Bacik path->slots[0]--; 3591aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3592aee68ee5SJosef Bacik path->slots[0]); 35933173a18fSJosef Bacik if (key.objectid == bytenr && 3594aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3595aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3596aee68ee5SJosef Bacik key.offset == blocksize))) 35973173a18fSJosef Bacik ret = 0; 35983173a18fSJosef Bacik } 3599aee68ee5SJosef Bacik 3600aee68ee5SJosef Bacik if (ret) { 3601aee68ee5SJosef Bacik skinny = false; 3602aee68ee5SJosef Bacik btrfs_release_path(path); 3603aee68ee5SJosef Bacik goto again; 3604aee68ee5SJosef Bacik } 3605aee68ee5SJosef Bacik } 3606cdccee99SLiu Bo if (ret) { 3607cdccee99SLiu Bo ASSERT(ret == 1); 3608cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3609cdccee99SLiu Bo btrfs_err(fs_info, 3610cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3611cdccee99SLiu Bo bytenr); 3612cdccee99SLiu Bo WARN_ON(1); 3613cdccee99SLiu Bo ret = -EINVAL; 3614cdccee99SLiu Bo goto out; 3615cdccee99SLiu Bo } 36163173a18fSJosef Bacik 36175d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 36185d4f98a2SYan Zheng out: 36195d4f98a2SYan Zheng btrfs_free_path(path); 36205d4f98a2SYan Zheng return ret; 36215d4f98a2SYan Zheng } 36225d4f98a2SYan Zheng 36235d4f98a2SYan Zheng /* 36245d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 36255d4f98a2SYan Zheng */ 36265d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 36275d4f98a2SYan Zheng struct extent_buffer *eb) 36285d4f98a2SYan Zheng { 36295d4f98a2SYan Zheng u64 flags; 36305d4f98a2SYan Zheng int ret; 36315d4f98a2SYan Zheng 36325d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 36335d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 36345d4f98a2SYan Zheng return 1; 36355d4f98a2SYan Zheng 36362ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info, 36373173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 36383173a18fSJosef Bacik NULL, &flags); 36395d4f98a2SYan Zheng BUG_ON(ret); 36405d4f98a2SYan Zheng 36415d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 36425d4f98a2SYan Zheng ret = 1; 36435d4f98a2SYan Zheng else 36445d4f98a2SYan Zheng ret = 0; 36455d4f98a2SYan Zheng return ret; 36465d4f98a2SYan Zheng } 36475d4f98a2SYan Zheng 36480af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 364932da5386SDavid Sterba struct btrfs_block_group *block_group, 36501bbc621eSChris Mason struct inode *inode, 36511bbc621eSChris Mason u64 ino) 36520af3d00bSJosef Bacik { 36530af3d00bSJosef Bacik struct btrfs_key key; 36540af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 36550af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 36560af3d00bSJosef Bacik int ret = 0; 36570af3d00bSJosef Bacik 36580af3d00bSJosef Bacik if (inode) 36590af3d00bSJosef Bacik goto truncate; 36600af3d00bSJosef Bacik 36610af3d00bSJosef Bacik key.objectid = ino; 36620af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 36630af3d00bSJosef Bacik key.offset = 0; 36640af3d00bSJosef Bacik 36654c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 36662e19f1f9SAl Viro if (IS_ERR(inode)) 36670af3d00bSJosef Bacik return -ENOENT; 36680af3d00bSJosef Bacik 36690af3d00bSJosef Bacik truncate: 36702ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 36717b61cd92SMiao Xie &fs_info->global_block_rsv); 36727b61cd92SMiao Xie if (ret) 36737b61cd92SMiao Xie goto out; 36747b61cd92SMiao Xie 36757a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 36760af3d00bSJosef Bacik if (IS_ERR(trans)) { 36773612b495STsutomu Itoh ret = PTR_ERR(trans); 36780af3d00bSJosef Bacik goto out; 36790af3d00bSJosef Bacik } 36800af3d00bSJosef Bacik 368177ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 36820af3d00bSJosef Bacik 36833a45bb20SJeff Mahoney btrfs_end_transaction(trans); 36842ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36850af3d00bSJosef Bacik out: 36860af3d00bSJosef Bacik iput(inode); 36870af3d00bSJosef Bacik return ret; 36880af3d00bSJosef Bacik } 36890af3d00bSJosef Bacik 36905d4f98a2SYan Zheng /* 36915d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 36925d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 36935d4f98a2SYan Zheng */ 36945d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 36955d4f98a2SYan Zheng struct btrfs_key *extent_key, 36965d4f98a2SYan Zheng struct extent_buffer *leaf, 36975d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 36985d4f98a2SYan Zheng struct rb_root *blocks) 36995d4f98a2SYan Zheng { 37000b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 37015d4f98a2SYan Zheng struct btrfs_path *path; 37025d4f98a2SYan Zheng struct tree_block *block; 37035d4f98a2SYan Zheng struct btrfs_root *root; 37045d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 37055d4f98a2SYan Zheng struct rb_node *rb_node; 37065d4f98a2SYan Zheng struct btrfs_key key; 37075d4f98a2SYan Zheng u64 ref_root; 37085d4f98a2SYan Zheng u64 ref_objectid; 37095d4f98a2SYan Zheng u64 ref_offset; 37105d4f98a2SYan Zheng u32 ref_count; 37115d4f98a2SYan Zheng u32 nritems; 37125d4f98a2SYan Zheng int err = 0; 37135d4f98a2SYan Zheng int added = 0; 37145d4f98a2SYan Zheng int counted; 37155d4f98a2SYan Zheng int ret; 37165d4f98a2SYan Zheng 37175d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 37185d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 37195d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 37205d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 37215d4f98a2SYan Zheng 37220af3d00bSJosef Bacik /* 37230af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 37240af3d00bSJosef Bacik * it and redo the search. 37250af3d00bSJosef Bacik */ 37260af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 37270b246afaSJeff Mahoney ret = delete_block_group_cache(fs_info, rc->block_group, 37280af3d00bSJosef Bacik NULL, ref_objectid); 37290af3d00bSJosef Bacik if (ret != -ENOENT) 37300af3d00bSJosef Bacik return ret; 37310af3d00bSJosef Bacik ret = 0; 37320af3d00bSJosef Bacik } 37330af3d00bSJosef Bacik 37340af3d00bSJosef Bacik path = btrfs_alloc_path(); 37350af3d00bSJosef Bacik if (!path) 37360af3d00bSJosef Bacik return -ENOMEM; 3737e4058b54SDavid Sterba path->reada = READA_FORWARD; 37380af3d00bSJosef Bacik 37390b246afaSJeff Mahoney root = read_fs_root(fs_info, ref_root); 37405d4f98a2SYan Zheng if (IS_ERR(root)) { 37415d4f98a2SYan Zheng err = PTR_ERR(root); 37423d7babdcSJosef Bacik goto out_free; 37433d7babdcSJosef Bacik } 37445d4f98a2SYan Zheng 37455d4f98a2SYan Zheng key.objectid = ref_objectid; 37465d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 374784850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 374884850e8dSYan, Zheng key.offset = 0; 374984850e8dSYan, Zheng else 375084850e8dSYan, Zheng key.offset = ref_offset; 37515d4f98a2SYan Zheng 37525d4f98a2SYan Zheng path->search_commit_root = 1; 37535d4f98a2SYan Zheng path->skip_locking = 1; 37545d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 37555d4f98a2SYan Zheng if (ret < 0) { 37565d4f98a2SYan Zheng err = ret; 37575d4f98a2SYan Zheng goto out; 37585d4f98a2SYan Zheng } 37595d4f98a2SYan Zheng 37605d4f98a2SYan Zheng leaf = path->nodes[0]; 37615d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 37625d4f98a2SYan Zheng /* 37635d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 37645d4f98a2SYan Zheng * are not counted in 37655d4f98a2SYan Zheng */ 37665d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 37675d4f98a2SYan Zheng counted = 0; 37685d4f98a2SYan Zheng else 37695d4f98a2SYan Zheng counted = 1; 37705d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 37715d4f98a2SYan Zheng if (rb_node) { 37725d4f98a2SYan Zheng if (counted) 37735d4f98a2SYan Zheng added = 1; 37745d4f98a2SYan Zheng else 37755d4f98a2SYan Zheng path->slots[0] = nritems; 37765d4f98a2SYan Zheng } 37775d4f98a2SYan Zheng 37785d4f98a2SYan Zheng while (ref_count > 0) { 37795d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 37805d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 37815d4f98a2SYan Zheng if (ret < 0) { 37825d4f98a2SYan Zheng err = ret; 37835d4f98a2SYan Zheng goto out; 37845d4f98a2SYan Zheng } 3785fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 37865d4f98a2SYan Zheng goto out; 37875d4f98a2SYan Zheng 37885d4f98a2SYan Zheng leaf = path->nodes[0]; 37895d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 37905d4f98a2SYan Zheng added = 0; 37915d4f98a2SYan Zheng 37925d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 37935d4f98a2SYan Zheng counted = 0; 37945d4f98a2SYan Zheng else 37955d4f98a2SYan Zheng counted = 1; 37965d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 37975d4f98a2SYan Zheng if (rb_node) { 37985d4f98a2SYan Zheng if (counted) 37995d4f98a2SYan Zheng added = 1; 38005d4f98a2SYan Zheng else 38015d4f98a2SYan Zheng path->slots[0] = nritems; 38025d4f98a2SYan Zheng } 38035d4f98a2SYan Zheng } 38045d4f98a2SYan Zheng 38055d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3806fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3807fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 38085d4f98a2SYan Zheng break; 38095d4f98a2SYan Zheng 38105d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 38115d4f98a2SYan Zheng struct btrfs_file_extent_item); 38125d4f98a2SYan Zheng 38135d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 38145d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 38155d4f98a2SYan Zheng goto next; 38165d4f98a2SYan Zheng 38175d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 38185d4f98a2SYan Zheng extent_key->objectid) 38195d4f98a2SYan Zheng goto next; 38205d4f98a2SYan Zheng 38215d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 38225d4f98a2SYan Zheng if (key.offset != ref_offset) 38235d4f98a2SYan Zheng goto next; 38245d4f98a2SYan Zheng 38255d4f98a2SYan Zheng if (counted) 38265d4f98a2SYan Zheng ref_count--; 38275d4f98a2SYan Zheng if (added) 38285d4f98a2SYan Zheng goto next; 38295d4f98a2SYan Zheng 38307476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 38315d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 38325d4f98a2SYan Zheng if (!block) { 38335d4f98a2SYan Zheng err = -ENOMEM; 38345d4f98a2SYan Zheng break; 38355d4f98a2SYan Zheng } 38365d4f98a2SYan Zheng block->bytenr = leaf->start; 38375d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 38385d4f98a2SYan Zheng block->level = 0; 38395d4f98a2SYan Zheng block->key_ready = 1; 38405d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 38415d4f98a2SYan Zheng &block->rb_node); 384243c04fb1SJeff Mahoney if (rb_node) 384343c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 384443c04fb1SJeff Mahoney block->bytenr); 38455d4f98a2SYan Zheng } 38465d4f98a2SYan Zheng if (counted) 38475d4f98a2SYan Zheng added = 1; 38485d4f98a2SYan Zheng else 38495d4f98a2SYan Zheng path->slots[0] = nritems; 38505d4f98a2SYan Zheng next: 38515d4f98a2SYan Zheng path->slots[0]++; 38525d4f98a2SYan Zheng 38535d4f98a2SYan Zheng } 38545d4f98a2SYan Zheng out: 385500246528SJosef Bacik btrfs_put_root(root); 38563d7babdcSJosef Bacik out_free: 38575d4f98a2SYan Zheng btrfs_free_path(path); 38585d4f98a2SYan Zheng return err; 38595d4f98a2SYan Zheng } 38605d4f98a2SYan Zheng 38615d4f98a2SYan Zheng /* 38622c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 38635d4f98a2SYan Zheng */ 38645d4f98a2SYan Zheng static noinline_for_stack 38655d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 38665d4f98a2SYan Zheng struct btrfs_key *extent_key, 38675d4f98a2SYan Zheng struct btrfs_path *path, 38685d4f98a2SYan Zheng struct rb_root *blocks) 38695d4f98a2SYan Zheng { 38705d4f98a2SYan Zheng struct btrfs_key key; 38715d4f98a2SYan Zheng struct extent_buffer *eb; 38725d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 38735d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 38745d4f98a2SYan Zheng unsigned long ptr; 38755d4f98a2SYan Zheng unsigned long end; 3876da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 3877647f63bdSFilipe David Borba Manana int ret = 0; 38785d4f98a2SYan Zheng int err = 0; 38795d4f98a2SYan Zheng 38805d4f98a2SYan Zheng eb = path->nodes[0]; 38815d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 38825d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 38835d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 38845d4f98a2SYan Zheng 38855d4f98a2SYan Zheng while (ptr < end) { 38865d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 38873de28d57SLiu Bo key.type = btrfs_get_extent_inline_ref_type(eb, iref, 38883de28d57SLiu Bo BTRFS_REF_TYPE_DATA); 38895d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 38905d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 38915d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 38925d4f98a2SYan Zheng blocks); 38935d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 38945d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 38955d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 38965d4f98a2SYan Zheng eb, dref, blocks); 38975d4f98a2SYan Zheng } else { 3898af431dcbSSu Yue ret = -EUCLEAN; 3899b14c55a1SLiu Bo btrfs_err(rc->extent_root->fs_info, 3900b14c55a1SLiu Bo "extent %llu slot %d has an invalid inline ref type", 3901b14c55a1SLiu Bo eb->start, path->slots[0]); 39025d4f98a2SYan Zheng } 3903647f63bdSFilipe David Borba Manana if (ret) { 3904647f63bdSFilipe David Borba Manana err = ret; 3905647f63bdSFilipe David Borba Manana goto out; 3906647f63bdSFilipe David Borba Manana } 39075d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 39085d4f98a2SYan Zheng } 39095d4f98a2SYan Zheng WARN_ON(ptr > end); 39105d4f98a2SYan Zheng 39115d4f98a2SYan Zheng while (1) { 39125d4f98a2SYan Zheng cond_resched(); 39135d4f98a2SYan Zheng eb = path->nodes[0]; 39145d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 39155d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 39165d4f98a2SYan Zheng if (ret < 0) { 39175d4f98a2SYan Zheng err = ret; 39185d4f98a2SYan Zheng break; 39195d4f98a2SYan Zheng } 39205d4f98a2SYan Zheng if (ret > 0) 39215d4f98a2SYan Zheng break; 39225d4f98a2SYan Zheng eb = path->nodes[0]; 39235d4f98a2SYan Zheng } 39245d4f98a2SYan Zheng 39255d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 39265d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 39275d4f98a2SYan Zheng break; 39285d4f98a2SYan Zheng 39295d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 39305d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 39315d4f98a2SYan Zheng blocks); 39325d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 39335d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 39345d4f98a2SYan Zheng struct btrfs_extent_data_ref); 39355d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 39365d4f98a2SYan Zheng eb, dref, blocks); 39376d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 3938ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3939ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3940ba3c2b19SNikolay Borisov ret = -EINVAL; 39415d4f98a2SYan Zheng } else { 39425d4f98a2SYan Zheng ret = 0; 39435d4f98a2SYan Zheng } 39445d4f98a2SYan Zheng if (ret) { 39455d4f98a2SYan Zheng err = ret; 39465d4f98a2SYan Zheng break; 39475d4f98a2SYan Zheng } 39485d4f98a2SYan Zheng path->slots[0]++; 39495d4f98a2SYan Zheng } 3950647f63bdSFilipe David Borba Manana out: 3951b3b4aa74SDavid Sterba btrfs_release_path(path); 39525d4f98a2SYan Zheng if (err) 39535d4f98a2SYan Zheng free_block_list(blocks); 39545d4f98a2SYan Zheng return err; 39555d4f98a2SYan Zheng } 39565d4f98a2SYan Zheng 39575d4f98a2SYan Zheng /* 39582c016dc2SLiu Bo * helper to find next unprocessed extent 39595d4f98a2SYan Zheng */ 39605d4f98a2SYan Zheng static noinline_for_stack 3961147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 39623fd0a558SYan, Zheng struct btrfs_key *extent_key) 39635d4f98a2SYan Zheng { 39640b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39655d4f98a2SYan Zheng struct btrfs_key key; 39665d4f98a2SYan Zheng struct extent_buffer *leaf; 39675d4f98a2SYan Zheng u64 start, end, last; 39685d4f98a2SYan Zheng int ret; 39695d4f98a2SYan Zheng 3970b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 39715d4f98a2SYan Zheng while (1) { 39725d4f98a2SYan Zheng cond_resched(); 39735d4f98a2SYan Zheng if (rc->search_start >= last) { 39745d4f98a2SYan Zheng ret = 1; 39755d4f98a2SYan Zheng break; 39765d4f98a2SYan Zheng } 39775d4f98a2SYan Zheng 39785d4f98a2SYan Zheng key.objectid = rc->search_start; 39795d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 39805d4f98a2SYan Zheng key.offset = 0; 39815d4f98a2SYan Zheng 39825d4f98a2SYan Zheng path->search_commit_root = 1; 39835d4f98a2SYan Zheng path->skip_locking = 1; 39845d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 39855d4f98a2SYan Zheng 0, 0); 39865d4f98a2SYan Zheng if (ret < 0) 39875d4f98a2SYan Zheng break; 39885d4f98a2SYan Zheng next: 39895d4f98a2SYan Zheng leaf = path->nodes[0]; 39905d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 39915d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 39925d4f98a2SYan Zheng if (ret != 0) 39935d4f98a2SYan Zheng break; 39945d4f98a2SYan Zheng leaf = path->nodes[0]; 39955d4f98a2SYan Zheng } 39965d4f98a2SYan Zheng 39975d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 39985d4f98a2SYan Zheng if (key.objectid >= last) { 39995d4f98a2SYan Zheng ret = 1; 40005d4f98a2SYan Zheng break; 40015d4f98a2SYan Zheng } 40025d4f98a2SYan Zheng 40033173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 40043173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 40053173a18fSJosef Bacik path->slots[0]++; 40063173a18fSJosef Bacik goto next; 40073173a18fSJosef Bacik } 40083173a18fSJosef Bacik 40093173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 40105d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 40115d4f98a2SYan Zheng path->slots[0]++; 40125d4f98a2SYan Zheng goto next; 40135d4f98a2SYan Zheng } 40145d4f98a2SYan Zheng 40153173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 40160b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 40173173a18fSJosef Bacik rc->search_start) { 40183173a18fSJosef Bacik path->slots[0]++; 40193173a18fSJosef Bacik goto next; 40203173a18fSJosef Bacik } 40213173a18fSJosef Bacik 40225d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 40235d4f98a2SYan Zheng key.objectid, &start, &end, 4024e6138876SJosef Bacik EXTENT_DIRTY, NULL); 40255d4f98a2SYan Zheng 40265d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 4027b3b4aa74SDavid Sterba btrfs_release_path(path); 40285d4f98a2SYan Zheng rc->search_start = end + 1; 40295d4f98a2SYan Zheng } else { 40303173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 40315d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 40323173a18fSJosef Bacik else 40333173a18fSJosef Bacik rc->search_start = key.objectid + 40340b246afaSJeff Mahoney fs_info->nodesize; 40353fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 40365d4f98a2SYan Zheng return 0; 40375d4f98a2SYan Zheng } 40385d4f98a2SYan Zheng } 4039b3b4aa74SDavid Sterba btrfs_release_path(path); 40405d4f98a2SYan Zheng return ret; 40415d4f98a2SYan Zheng } 40425d4f98a2SYan Zheng 40435d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 40445d4f98a2SYan Zheng { 40455d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 40467585717fSChris Mason 40477585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 40485d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 40497585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 40505d4f98a2SYan Zheng } 40515d4f98a2SYan Zheng 40525d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 40535d4f98a2SYan Zheng { 40545d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 40557585717fSChris Mason 40567585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 40575d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 40587585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 40595d4f98a2SYan Zheng } 40605d4f98a2SYan Zheng 40615d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 40625d4f98a2SYan Zheng { 40635d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 40645d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 40655d4f98a2SYan Zheng return 1; 40665d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 40675d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 40685d4f98a2SYan Zheng return 1; 40695d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 40705d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 40715d4f98a2SYan Zheng return 1; 40725d4f98a2SYan Zheng return 0; 40735d4f98a2SYan Zheng } 40745d4f98a2SYan Zheng 40753fd0a558SYan, Zheng static noinline_for_stack 40763fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 40773fd0a558SYan, Zheng { 40783fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 4079ac2fabacSJosef Bacik int ret; 40803fd0a558SYan, Zheng 40812ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 408266d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 40833fd0a558SYan, Zheng if (!rc->block_rsv) 40843fd0a558SYan, Zheng return -ENOMEM; 40853fd0a558SYan, Zheng 40863fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 4087b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 40883fd0a558SYan, Zheng rc->extents_found = 0; 40893fd0a558SYan, Zheng rc->nodes_relocated = 0; 40903fd0a558SYan, Zheng rc->merging_rsv_size = 0; 40910647bf56SWang Shilong rc->reserved_bytes = 0; 4092da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 40930647bf56SWang Shilong RELOCATION_RESERVED_NODES; 4094ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 4095ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 4096ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 4097ac2fabacSJosef Bacik if (ret) 4098ac2fabacSJosef Bacik return ret; 40993fd0a558SYan, Zheng 41003fd0a558SYan, Zheng rc->create_reloc_tree = 1; 41013fd0a558SYan, Zheng set_reloc_control(rc); 41023fd0a558SYan, Zheng 41037a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 410428818947SLiu Bo if (IS_ERR(trans)) { 410528818947SLiu Bo unset_reloc_control(rc); 410628818947SLiu Bo /* 410728818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 410828818947SLiu Bo * cleanup. And callers are responsible to free the above 410928818947SLiu Bo * block rsv. 411028818947SLiu Bo */ 411128818947SLiu Bo return PTR_ERR(trans); 411228818947SLiu Bo } 41133a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 41143fd0a558SYan, Zheng return 0; 41153fd0a558SYan, Zheng } 411676dda93cSYan, Zheng 41175d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 41185d4f98a2SYan Zheng { 41192ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 41205d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 41215d4f98a2SYan Zheng struct btrfs_key key; 41225d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 41235d4f98a2SYan Zheng struct btrfs_path *path; 41245d4f98a2SYan Zheng struct btrfs_extent_item *ei; 41255d4f98a2SYan Zheng u64 flags; 41265d4f98a2SYan Zheng u32 item_size; 41275d4f98a2SYan Zheng int ret; 41285d4f98a2SYan Zheng int err = 0; 4129c87f08caSChris Mason int progress = 0; 41305d4f98a2SYan Zheng 41315d4f98a2SYan Zheng path = btrfs_alloc_path(); 41323fd0a558SYan, Zheng if (!path) 41335d4f98a2SYan Zheng return -ENOMEM; 4134e4058b54SDavid Sterba path->reada = READA_FORWARD; 41353fd0a558SYan, Zheng 41363fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 41373fd0a558SYan, Zheng if (ret) { 41383fd0a558SYan, Zheng err = ret; 41393fd0a558SYan, Zheng goto out_free; 41402423fdfbSJiri Slaby } 41415d4f98a2SYan Zheng 41425d4f98a2SYan Zheng while (1) { 41430647bf56SWang Shilong rc->reserved_bytes = 0; 41440647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 41450647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 41460647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 41470647bf56SWang Shilong if (ret) { 41480647bf56SWang Shilong err = ret; 41490647bf56SWang Shilong break; 41500647bf56SWang Shilong } 4151c87f08caSChris Mason progress++; 4152a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 41530f788c58SLiu Bo if (IS_ERR(trans)) { 41540f788c58SLiu Bo err = PTR_ERR(trans); 41550f788c58SLiu Bo trans = NULL; 41560f788c58SLiu Bo break; 41570f788c58SLiu Bo } 4158c87f08caSChris Mason restart: 41593fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 41603a45bb20SJeff Mahoney btrfs_end_transaction(trans); 416142a657f5SPan Bian trans = NULL; 41623fd0a558SYan, Zheng continue; 41633fd0a558SYan, Zheng } 41643fd0a558SYan, Zheng 4165147d256eSZhaolei ret = find_next_extent(rc, path, &key); 41665d4f98a2SYan Zheng if (ret < 0) 41675d4f98a2SYan Zheng err = ret; 41685d4f98a2SYan Zheng if (ret != 0) 41695d4f98a2SYan Zheng break; 41705d4f98a2SYan Zheng 41715d4f98a2SYan Zheng rc->extents_found++; 41725d4f98a2SYan Zheng 41735d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 41745d4f98a2SYan Zheng struct btrfs_extent_item); 41753fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 41765d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 41775d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 41785d4f98a2SYan Zheng ret = check_extent_flags(flags); 41795d4f98a2SYan Zheng BUG_ON(ret); 41806d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 4181ba3c2b19SNikolay Borisov err = -EINVAL; 4182ba3c2b19SNikolay Borisov btrfs_print_v0_err(trans->fs_info); 4183ba3c2b19SNikolay Borisov btrfs_abort_transaction(trans, err); 4184ba3c2b19SNikolay Borisov break; 41855d4f98a2SYan Zheng } else { 41865d4f98a2SYan Zheng BUG(); 41875d4f98a2SYan Zheng } 41885d4f98a2SYan Zheng 41895d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 41905d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 41915d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 41925d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41935d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 41945d4f98a2SYan Zheng } else { 4195b3b4aa74SDavid Sterba btrfs_release_path(path); 41965d4f98a2SYan Zheng ret = 0; 41975d4f98a2SYan Zheng } 41985d4f98a2SYan Zheng if (ret < 0) { 41993fd0a558SYan, Zheng err = ret; 42005d4f98a2SYan Zheng break; 42015d4f98a2SYan Zheng } 42025d4f98a2SYan Zheng 42035d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 42045d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 42055d4f98a2SYan Zheng if (ret < 0) { 42061708cc57SWang Shilong /* 42071708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 42081708cc57SWang Shilong * backref cache when committing transaction. 42091708cc57SWang Shilong */ 42101708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 42111708cc57SWang Shilong 42123fd0a558SYan, Zheng if (ret != -EAGAIN) { 42135d4f98a2SYan Zheng err = ret; 42145d4f98a2SYan Zheng break; 42155d4f98a2SYan Zheng } 42163fd0a558SYan, Zheng rc->extents_found--; 42173fd0a558SYan, Zheng rc->search_start = key.objectid; 42183fd0a558SYan, Zheng } 42195d4f98a2SYan Zheng } 42205d4f98a2SYan Zheng 42213a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 42222ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42233fd0a558SYan, Zheng trans = NULL; 42245d4f98a2SYan Zheng 42255d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 42265d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 42275d4f98a2SYan Zheng rc->found_file_extent = 1; 42280257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 42293fd0a558SYan, Zheng &key, &rc->cluster); 42305d4f98a2SYan Zheng if (ret < 0) { 42315d4f98a2SYan Zheng err = ret; 42325d4f98a2SYan Zheng break; 42335d4f98a2SYan Zheng } 42345d4f98a2SYan Zheng } 4235f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 4236f31ea088SQu Wenruo err = -ECANCELED; 4237f31ea088SQu Wenruo break; 4238f31ea088SQu Wenruo } 42395d4f98a2SYan Zheng } 4240c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 424143a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 42429689457bSShilong Wang if (ret == 1) { 4243c87f08caSChris Mason err = 0; 4244c87f08caSChris Mason progress = 0; 4245c87f08caSChris Mason goto restart; 4246c87f08caSChris Mason } 4247c87f08caSChris Mason } 42483fd0a558SYan, Zheng 4249b3b4aa74SDavid Sterba btrfs_release_path(path); 425091166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 42515d4f98a2SYan Zheng 42525d4f98a2SYan Zheng if (trans) { 42533a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 42542ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42555d4f98a2SYan Zheng } 42565d4f98a2SYan Zheng 42570257bb82SYan, Zheng if (!err) { 42583fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 42593fd0a558SYan, Zheng &rc->cluster); 42600257bb82SYan, Zheng if (ret < 0) 42610257bb82SYan, Zheng err = ret; 42620257bb82SYan, Zheng } 42630257bb82SYan, Zheng 42643fd0a558SYan, Zheng rc->create_reloc_tree = 0; 42653fd0a558SYan, Zheng set_reloc_control(rc); 42660257bb82SYan, Zheng 42673fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 4268*63f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 42695d4f98a2SYan Zheng 42707f913c7cSQu Wenruo /* 42717f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 42727f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 42737f913c7cSQu Wenruo * 42747f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 42757f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 42767f913c7cSQu Wenruo * merge_reloc_roots() 42777f913c7cSQu Wenruo */ 42783fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 42795d4f98a2SYan Zheng 42805d4f98a2SYan Zheng merge_reloc_roots(rc); 42815d4f98a2SYan Zheng 42823fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 42835d4f98a2SYan Zheng unset_reloc_control(rc); 4284*63f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 42855d4f98a2SYan Zheng 42865d4f98a2SYan Zheng /* get rid of pinned extents */ 42877a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 428862b99540SQu Wenruo if (IS_ERR(trans)) { 42893612b495STsutomu Itoh err = PTR_ERR(trans); 429062b99540SQu Wenruo goto out_free; 429162b99540SQu Wenruo } 42923a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 4293d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4294d2311e69SQu Wenruo if (ret < 0 && !err) 4295d2311e69SQu Wenruo err = ret; 42963fd0a558SYan, Zheng out_free: 42972ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 42983fd0a558SYan, Zheng btrfs_free_path(path); 42995d4f98a2SYan Zheng return err; 43005d4f98a2SYan Zheng } 43015d4f98a2SYan Zheng 43025d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 43030257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 43045d4f98a2SYan Zheng { 43055d4f98a2SYan Zheng struct btrfs_path *path; 43065d4f98a2SYan Zheng struct btrfs_inode_item *item; 43075d4f98a2SYan Zheng struct extent_buffer *leaf; 43085d4f98a2SYan Zheng int ret; 43095d4f98a2SYan Zheng 43105d4f98a2SYan Zheng path = btrfs_alloc_path(); 43115d4f98a2SYan Zheng if (!path) 43125d4f98a2SYan Zheng return -ENOMEM; 43135d4f98a2SYan Zheng 43145d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 43155d4f98a2SYan Zheng if (ret) 43165d4f98a2SYan Zheng goto out; 43175d4f98a2SYan Zheng 43185d4f98a2SYan Zheng leaf = path->nodes[0]; 43195d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4320b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 43215d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 43220257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 43235d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 43243fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 43253fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 43265d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 43275d4f98a2SYan Zheng out: 43285d4f98a2SYan Zheng btrfs_free_path(path); 43295d4f98a2SYan Zheng return ret; 43305d4f98a2SYan Zheng } 43315d4f98a2SYan Zheng 43325d4f98a2SYan Zheng /* 43335d4f98a2SYan Zheng * helper to create inode for data relocation. 43345d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 43355d4f98a2SYan Zheng */ 43363fd0a558SYan, Zheng static noinline_for_stack 43373fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 433832da5386SDavid Sterba struct btrfs_block_group *group) 43395d4f98a2SYan Zheng { 43405d4f98a2SYan Zheng struct inode *inode = NULL; 43415d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 43425d4f98a2SYan Zheng struct btrfs_root *root; 43435d4f98a2SYan Zheng struct btrfs_key key; 43444624900dSZhaolei u64 objectid; 43455d4f98a2SYan Zheng int err = 0; 43465d4f98a2SYan Zheng 43475d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 43485d4f98a2SYan Zheng if (IS_ERR(root)) 43495d4f98a2SYan Zheng return ERR_CAST(root); 43505d4f98a2SYan Zheng 4351a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 435276deacf0SJosef Bacik if (IS_ERR(trans)) { 435300246528SJosef Bacik btrfs_put_root(root); 43543fd0a558SYan, Zheng return ERR_CAST(trans); 435576deacf0SJosef Bacik } 43565d4f98a2SYan Zheng 4357581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 43585d4f98a2SYan Zheng if (err) 43595d4f98a2SYan Zheng goto out; 43605d4f98a2SYan Zheng 43610257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 43625d4f98a2SYan Zheng BUG_ON(err); 43635d4f98a2SYan Zheng 43645d4f98a2SYan Zheng key.objectid = objectid; 43655d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 43665d4f98a2SYan Zheng key.offset = 0; 43674c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 43682e19f1f9SAl Viro BUG_ON(IS_ERR(inode)); 4369b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 43705d4f98a2SYan Zheng 437173f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 43725d4f98a2SYan Zheng out: 437300246528SJosef Bacik btrfs_put_root(root); 43743a45bb20SJeff Mahoney btrfs_end_transaction(trans); 43752ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 43765d4f98a2SYan Zheng if (err) { 43775d4f98a2SYan Zheng if (inode) 43785d4f98a2SYan Zheng iput(inode); 43795d4f98a2SYan Zheng inode = ERR_PTR(err); 43805d4f98a2SYan Zheng } 43815d4f98a2SYan Zheng return inode; 43825d4f98a2SYan Zheng } 43835d4f98a2SYan Zheng 4384c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 43853fd0a558SYan, Zheng { 43863fd0a558SYan, Zheng struct reloc_control *rc; 43873fd0a558SYan, Zheng 43883fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 43893fd0a558SYan, Zheng if (!rc) 43903fd0a558SYan, Zheng return NULL; 43913fd0a558SYan, Zheng 43923fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 4393d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 43943fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 43953fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 439643eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 439743eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 43983fd0a558SYan, Zheng return rc; 43993fd0a558SYan, Zheng } 44003fd0a558SYan, Zheng 44015d4f98a2SYan Zheng /* 4402ebce0e01SAdam Borowski * Print the block group being relocated 4403ebce0e01SAdam Borowski */ 4404ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 440532da5386SDavid Sterba struct btrfs_block_group *block_group) 4406ebce0e01SAdam Borowski { 4407f89e09cfSAnand Jain char buf[128] = {'\0'}; 4408ebce0e01SAdam Borowski 4409f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4410ebce0e01SAdam Borowski 4411ebce0e01SAdam Borowski btrfs_info(fs_info, 4412ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4413b3470b5dSDavid Sterba block_group->start, buf); 4414ebce0e01SAdam Borowski } 4415ebce0e01SAdam Borowski 4416430640e3SQu Wenruo static const char *stage_to_string(int stage) 4417430640e3SQu Wenruo { 4418430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 4419430640e3SQu Wenruo return "move data extents"; 4420430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 4421430640e3SQu Wenruo return "update data pointers"; 4422430640e3SQu Wenruo return "unknown"; 4423430640e3SQu Wenruo } 4424430640e3SQu Wenruo 4425ebce0e01SAdam Borowski /* 44265d4f98a2SYan Zheng * function to relocate all extents in a block group. 44275d4f98a2SYan Zheng */ 44286bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 44295d4f98a2SYan Zheng { 443032da5386SDavid Sterba struct btrfs_block_group *bg; 44316bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 44325d4f98a2SYan Zheng struct reloc_control *rc; 44330af3d00bSJosef Bacik struct inode *inode; 44340af3d00bSJosef Bacik struct btrfs_path *path; 44355d4f98a2SYan Zheng int ret; 4436f0486c68SYan, Zheng int rw = 0; 44375d4f98a2SYan Zheng int err = 0; 44385d4f98a2SYan Zheng 4439eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4440eede2bf3SOmar Sandoval if (!bg) 4441eede2bf3SOmar Sandoval return -ENOENT; 4442eede2bf3SOmar Sandoval 4443eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4444eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4445eede2bf3SOmar Sandoval return -ETXTBSY; 4446eede2bf3SOmar Sandoval } 4447eede2bf3SOmar Sandoval 4448c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4449eede2bf3SOmar Sandoval if (!rc) { 4450eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 44515d4f98a2SYan Zheng return -ENOMEM; 4452eede2bf3SOmar Sandoval } 44535d4f98a2SYan Zheng 4454f0486c68SYan, Zheng rc->extent_root = extent_root; 4455eede2bf3SOmar Sandoval rc->block_group = bg; 44565d4f98a2SYan Zheng 4457b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4458f0486c68SYan, Zheng if (ret) { 4459f0486c68SYan, Zheng err = ret; 4460f0486c68SYan, Zheng goto out; 4461f0486c68SYan, Zheng } 4462f0486c68SYan, Zheng rw = 1; 4463f0486c68SYan, Zheng 44640af3d00bSJosef Bacik path = btrfs_alloc_path(); 44650af3d00bSJosef Bacik if (!path) { 44660af3d00bSJosef Bacik err = -ENOMEM; 44670af3d00bSJosef Bacik goto out; 44680af3d00bSJosef Bacik } 44690af3d00bSJosef Bacik 44707949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 44710af3d00bSJosef Bacik btrfs_free_path(path); 44720af3d00bSJosef Bacik 44730af3d00bSJosef Bacik if (!IS_ERR(inode)) 44741bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 44750af3d00bSJosef Bacik else 44760af3d00bSJosef Bacik ret = PTR_ERR(inode); 44770af3d00bSJosef Bacik 44780af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 44790af3d00bSJosef Bacik err = ret; 44800af3d00bSJosef Bacik goto out; 44810af3d00bSJosef Bacik } 44820af3d00bSJosef Bacik 44835d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 44845d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 44855d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 44865d4f98a2SYan Zheng rc->data_inode = NULL; 44875d4f98a2SYan Zheng goto out; 44885d4f98a2SYan Zheng } 44895d4f98a2SYan Zheng 44900b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 44915d4f98a2SYan Zheng 44929cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4493f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 44946374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4495b3470b5dSDavid Sterba rc->block_group->start, 4496b3470b5dSDavid Sterba rc->block_group->length); 44975d4f98a2SYan Zheng 44985d4f98a2SYan Zheng while (1) { 4499430640e3SQu Wenruo int finishes_stage; 4500430640e3SQu Wenruo 450176dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 45025d4f98a2SYan Zheng ret = relocate_block_group(rc); 450376dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4504ff612ba7SJosef Bacik if (ret < 0) 45055d4f98a2SYan Zheng err = ret; 4506ff612ba7SJosef Bacik 4507430640e3SQu Wenruo finishes_stage = rc->stage; 4508ff612ba7SJosef Bacik /* 4509ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4510ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4511ff612ba7SJosef Bacik * different block group we could end up hitting the 4512ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4513ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4514ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4515ff612ba7SJosef Bacik * out of the loop if we hit an error. 4516ff612ba7SJosef Bacik */ 4517ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4518ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4519ff612ba7SJosef Bacik (u64)-1); 4520ff612ba7SJosef Bacik if (ret) 4521ff612ba7SJosef Bacik err = ret; 4522ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4523ff612ba7SJosef Bacik 0, -1); 4524ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 45255d4f98a2SYan Zheng } 45265d4f98a2SYan Zheng 4527ff612ba7SJosef Bacik if (err < 0) 4528ff612ba7SJosef Bacik goto out; 4529ff612ba7SJosef Bacik 45305d4f98a2SYan Zheng if (rc->extents_found == 0) 45315d4f98a2SYan Zheng break; 45325d4f98a2SYan Zheng 4533430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4534430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 45355d4f98a2SYan Zheng } 45365d4f98a2SYan Zheng 45375d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 45385d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4539bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 45405d4f98a2SYan Zheng out: 4541f0486c68SYan, Zheng if (err && rw) 45422ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 45435d4f98a2SYan Zheng iput(rc->data_inode); 45445d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 45455d4f98a2SYan Zheng kfree(rc); 45465d4f98a2SYan Zheng return err; 45475d4f98a2SYan Zheng } 45485d4f98a2SYan Zheng 454976dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 455076dda93cSYan, Zheng { 45510b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 455276dda93cSYan, Zheng struct btrfs_trans_handle *trans; 455379787eaaSJeff Mahoney int ret, err; 455476dda93cSYan, Zheng 45550b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 455679787eaaSJeff Mahoney if (IS_ERR(trans)) 455779787eaaSJeff Mahoney return PTR_ERR(trans); 455876dda93cSYan, Zheng 455976dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 456076dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 456176dda93cSYan, Zheng root->root_item.drop_level = 0; 456276dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 45630b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 456476dda93cSYan, Zheng &root->root_key, &root->root_item); 456576dda93cSYan, Zheng 45663a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 456779787eaaSJeff Mahoney if (err) 456879787eaaSJeff Mahoney return err; 456979787eaaSJeff Mahoney return ret; 457076dda93cSYan, Zheng } 457176dda93cSYan, Zheng 45725d4f98a2SYan Zheng /* 45735d4f98a2SYan Zheng * recover relocation interrupted by system crash. 45745d4f98a2SYan Zheng * 45755d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 45765d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 45775d4f98a2SYan Zheng */ 45785d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 45795d4f98a2SYan Zheng { 45800b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 45815d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 45825d4f98a2SYan Zheng struct btrfs_key key; 45835d4f98a2SYan Zheng struct btrfs_root *fs_root; 45845d4f98a2SYan Zheng struct btrfs_root *reloc_root; 45855d4f98a2SYan Zheng struct btrfs_path *path; 45865d4f98a2SYan Zheng struct extent_buffer *leaf; 45875d4f98a2SYan Zheng struct reloc_control *rc = NULL; 45885d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 45895d4f98a2SYan Zheng int ret; 45905d4f98a2SYan Zheng int err = 0; 45915d4f98a2SYan Zheng 45925d4f98a2SYan Zheng path = btrfs_alloc_path(); 45935d4f98a2SYan Zheng if (!path) 45945d4f98a2SYan Zheng return -ENOMEM; 4595e4058b54SDavid Sterba path->reada = READA_BACK; 45965d4f98a2SYan Zheng 45975d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 45985d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 45995d4f98a2SYan Zheng key.offset = (u64)-1; 46005d4f98a2SYan Zheng 46015d4f98a2SYan Zheng while (1) { 46020b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 46035d4f98a2SYan Zheng path, 0, 0); 46045d4f98a2SYan Zheng if (ret < 0) { 46055d4f98a2SYan Zheng err = ret; 46065d4f98a2SYan Zheng goto out; 46075d4f98a2SYan Zheng } 46085d4f98a2SYan Zheng if (ret > 0) { 46095d4f98a2SYan Zheng if (path->slots[0] == 0) 46105d4f98a2SYan Zheng break; 46115d4f98a2SYan Zheng path->slots[0]--; 46125d4f98a2SYan Zheng } 46135d4f98a2SYan Zheng leaf = path->nodes[0]; 46145d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4615b3b4aa74SDavid Sterba btrfs_release_path(path); 46165d4f98a2SYan Zheng 46175d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 46185d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 46195d4f98a2SYan Zheng break; 46205d4f98a2SYan Zheng 46213dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(root, &key); 46225d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 46235d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 46245d4f98a2SYan Zheng goto out; 46255d4f98a2SYan Zheng } 46265d4f98a2SYan Zheng 46273dbf1738SJosef Bacik set_bit(BTRFS_ROOT_REF_COWS, &reloc_root->state); 46285d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 46295d4f98a2SYan Zheng 46305d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 46310b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 46325d4f98a2SYan Zheng reloc_root->root_key.offset); 46335d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 463476dda93cSYan, Zheng ret = PTR_ERR(fs_root); 463576dda93cSYan, Zheng if (ret != -ENOENT) { 463676dda93cSYan, Zheng err = ret; 46375d4f98a2SYan Zheng goto out; 46385d4f98a2SYan Zheng } 463979787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 464079787eaaSJeff Mahoney if (ret < 0) { 464179787eaaSJeff Mahoney err = ret; 464279787eaaSJeff Mahoney goto out; 464379787eaaSJeff Mahoney } 4644932fd26dSJosef Bacik } else { 464500246528SJosef Bacik btrfs_put_root(fs_root); 464676dda93cSYan, Zheng } 46475d4f98a2SYan Zheng } 46485d4f98a2SYan Zheng 46495d4f98a2SYan Zheng if (key.offset == 0) 46505d4f98a2SYan Zheng break; 46515d4f98a2SYan Zheng 46525d4f98a2SYan Zheng key.offset--; 46535d4f98a2SYan Zheng } 4654b3b4aa74SDavid Sterba btrfs_release_path(path); 46555d4f98a2SYan Zheng 46565d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 46575d4f98a2SYan Zheng goto out; 46585d4f98a2SYan Zheng 4659c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 46605d4f98a2SYan Zheng if (!rc) { 46615d4f98a2SYan Zheng err = -ENOMEM; 46625d4f98a2SYan Zheng goto out; 46635d4f98a2SYan Zheng } 46645d4f98a2SYan Zheng 46650b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 46665d4f98a2SYan Zheng 46675d4f98a2SYan Zheng set_reloc_control(rc); 46685d4f98a2SYan Zheng 46697a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 46703612b495STsutomu Itoh if (IS_ERR(trans)) { 46713612b495STsutomu Itoh unset_reloc_control(rc); 46723612b495STsutomu Itoh err = PTR_ERR(trans); 46733612b495STsutomu Itoh goto out_free; 46743612b495STsutomu Itoh } 46753fd0a558SYan, Zheng 46763fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 46773fd0a558SYan, Zheng 46785d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 46795d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 46805d4f98a2SYan Zheng struct btrfs_root, root_list); 46815d4f98a2SYan Zheng list_del(&reloc_root->root_list); 46825d4f98a2SYan Zheng 46835d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 46845d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 46855d4f98a2SYan Zheng &rc->reloc_roots); 46865d4f98a2SYan Zheng continue; 46875d4f98a2SYan Zheng } 46885d4f98a2SYan Zheng 46890b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 469079787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 469179787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4692ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 469379787eaaSJeff Mahoney goto out_free; 469479787eaaSJeff Mahoney } 46955d4f98a2SYan Zheng 4696ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 469779787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 46985d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 469900246528SJosef Bacik btrfs_put_root(fs_root); 47005d4f98a2SYan Zheng } 47015d4f98a2SYan Zheng 47023a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 470379787eaaSJeff Mahoney if (err) 470479787eaaSJeff Mahoney goto out_free; 47055d4f98a2SYan Zheng 47065d4f98a2SYan Zheng merge_reloc_roots(rc); 47075d4f98a2SYan Zheng 47085d4f98a2SYan Zheng unset_reloc_control(rc); 47095d4f98a2SYan Zheng 47107a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 471162b99540SQu Wenruo if (IS_ERR(trans)) { 47123612b495STsutomu Itoh err = PTR_ERR(trans); 471362b99540SQu Wenruo goto out_free; 471462b99540SQu Wenruo } 47153a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 4716d2311e69SQu Wenruo 4717d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4718d2311e69SQu Wenruo if (ret < 0 && !err) 4719d2311e69SQu Wenruo err = ret; 47203612b495STsutomu Itoh out_free: 47215d4f98a2SYan Zheng kfree(rc); 47223612b495STsutomu Itoh out: 4723aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4724aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4725aca1bba6SLiu Bo 47265d4f98a2SYan Zheng btrfs_free_path(path); 47275d4f98a2SYan Zheng 47285d4f98a2SYan Zheng if (err == 0) { 47295d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 47300b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 4731932fd26dSJosef Bacik if (IS_ERR(fs_root)) { 47325d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4733932fd26dSJosef Bacik } else { 473466b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 473500246528SJosef Bacik btrfs_put_root(fs_root); 4736932fd26dSJosef Bacik } 4737932fd26dSJosef Bacik } 47385d4f98a2SYan Zheng return err; 47395d4f98a2SYan Zheng } 47405d4f98a2SYan Zheng 47415d4f98a2SYan Zheng /* 47425d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 47435d4f98a2SYan Zheng * 47445d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 47455d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 47465d4f98a2SYan Zheng */ 47475d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 47485d4f98a2SYan Zheng { 47490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 47505d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 47515d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 47525d4f98a2SYan Zheng int ret; 47535d4f98a2SYan Zheng u64 disk_bytenr; 47544577b014SJosef Bacik u64 new_bytenr; 47555d4f98a2SYan Zheng LIST_HEAD(list); 47565d4f98a2SYan Zheng 47575d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4758bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 47595d4f98a2SYan Zheng 47605d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 47610b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4762a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 476379787eaaSJeff Mahoney if (ret) 476479787eaaSJeff Mahoney goto out; 47655d4f98a2SYan Zheng 47665d4f98a2SYan Zheng while (!list_empty(&list)) { 47675d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 47685d4f98a2SYan Zheng list_del_init(&sums->list); 47695d4f98a2SYan Zheng 47704577b014SJosef Bacik /* 47714577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 47724577b014SJosef Bacik * We need to do this because we will read in entire prealloc 47734577b014SJosef Bacik * extents but we may have written to say the middle of the 47744577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 47754577b014SJosef Bacik * the right disk offset. 47764577b014SJosef Bacik * 47774577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 47784577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 47794577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 47804577b014SJosef Bacik * disk length. 47814577b014SJosef Bacik */ 4782bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 47834577b014SJosef Bacik sums->bytenr = new_bytenr; 47845d4f98a2SYan Zheng 4785f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 47865d4f98a2SYan Zheng } 478779787eaaSJeff Mahoney out: 47885d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4789411fc6bcSAndi Kleen return ret; 47905d4f98a2SYan Zheng } 47913fd0a558SYan, Zheng 479283d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 47933fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 47943fd0a558SYan, Zheng struct extent_buffer *cow) 47953fd0a558SYan, Zheng { 47960b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 47973fd0a558SYan, Zheng struct reloc_control *rc; 47983fd0a558SYan, Zheng struct backref_node *node; 47993fd0a558SYan, Zheng int first_cow = 0; 48003fd0a558SYan, Zheng int level; 480183d4cfd4SJosef Bacik int ret = 0; 48023fd0a558SYan, Zheng 48030b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 48043fd0a558SYan, Zheng if (!rc) 480583d4cfd4SJosef Bacik return 0; 48063fd0a558SYan, Zheng 48073fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 48083fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 48093fd0a558SYan, Zheng 4810c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4811c974c464SWang Shilong if (buf == root->node) 4812c974c464SWang Shilong __update_reloc_root(root, cow->start); 4813c974c464SWang Shilong } 4814c974c464SWang Shilong 48153fd0a558SYan, Zheng level = btrfs_header_level(buf); 48163fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 48173fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 48183fd0a558SYan, Zheng first_cow = 1; 48193fd0a558SYan, Zheng 48203fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 48213fd0a558SYan, Zheng rc->create_reloc_tree) { 48223fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 48233fd0a558SYan, Zheng 48243fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 48253fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 48263fd0a558SYan, Zheng node->new_bytenr != buf->start); 48273fd0a558SYan, Zheng 48283fd0a558SYan, Zheng drop_node_buffer(node); 482967439dadSDavid Sterba atomic_inc(&cow->refs); 48303fd0a558SYan, Zheng node->eb = cow; 48313fd0a558SYan, Zheng node->new_bytenr = cow->start; 48323fd0a558SYan, Zheng 48333fd0a558SYan, Zheng if (!node->pending) { 48343fd0a558SYan, Zheng list_move_tail(&node->list, 48353fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 48363fd0a558SYan, Zheng node->pending = 1; 48373fd0a558SYan, Zheng } 48383fd0a558SYan, Zheng 48393fd0a558SYan, Zheng if (first_cow) 48403fd0a558SYan, Zheng __mark_block_processed(rc, node); 48413fd0a558SYan, Zheng 48423fd0a558SYan, Zheng if (first_cow && level > 0) 48433fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 48443fd0a558SYan, Zheng } 48453fd0a558SYan, Zheng 484683d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 48473fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 484883d4cfd4SJosef Bacik return ret; 48493fd0a558SYan, Zheng } 48503fd0a558SYan, Zheng 48513fd0a558SYan, Zheng /* 48523fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 485301327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 48543fd0a558SYan, Zheng */ 4855147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 48563fd0a558SYan, Zheng u64 *bytes_to_reserve) 48573fd0a558SYan, Zheng { 485810995c04SQu Wenruo struct btrfs_root *root = pending->root; 485910995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 48603fd0a558SYan, Zheng 48616282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 48623fd0a558SYan, Zheng return; 48633fd0a558SYan, Zheng 48643fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 48653fd0a558SYan, Zheng return; 48663fd0a558SYan, Zheng 48673fd0a558SYan, Zheng root = root->reloc_root; 48683fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 48693fd0a558SYan, Zheng /* 48703fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 48713fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 48723fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 48733fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 48743fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 48753fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 48763fd0a558SYan, Zheng * tree may use more space than it frees. so we need 48773fd0a558SYan, Zheng * reserve extra space. 48783fd0a558SYan, Zheng */ 48793fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 48803fd0a558SYan, Zheng } 48813fd0a558SYan, Zheng 48823fd0a558SYan, Zheng /* 48833fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 48843fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 48853fd0a558SYan, Zheng */ 488649b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 48873fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 48883fd0a558SYan, Zheng { 48893fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 48903fd0a558SYan, Zheng struct btrfs_root *reloc_root; 48913fd0a558SYan, Zheng struct btrfs_root *new_root; 489210995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 48933fd0a558SYan, Zheng int ret; 48943fd0a558SYan, Zheng 48956282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 489649b25e05SJeff Mahoney return 0; 48973fd0a558SYan, Zheng 48983fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 48993fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 49003fd0a558SYan, Zheng 49013fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 49023fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 49033fd0a558SYan, Zheng rc->block_rsv, 49043a584174SLu Fengqi rc->nodes_relocated, true); 490549b25e05SJeff Mahoney if (ret) 490649b25e05SJeff Mahoney return ret; 49073fd0a558SYan, Zheng } 49083fd0a558SYan, Zheng 49093fd0a558SYan, Zheng new_root = pending->snap; 49103fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 49113fd0a558SYan, Zheng new_root->root_key.objectid); 491249b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 491349b25e05SJeff Mahoney return PTR_ERR(reloc_root); 49143fd0a558SYan, Zheng 4915ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4916ffd7b339SJeff Mahoney BUG_ON(ret < 0); 49173fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 49183fd0a558SYan, Zheng 491949b25e05SJeff Mahoney if (rc->create_reloc_tree) 49203fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 492149b25e05SJeff Mahoney return ret; 49223fd0a558SYan, Zheng } 4923